You are on page 1of 458


Table of Contents

Authors' Biographies xv

Authors' Introductions and Guide to Study xvii

Introduction to First Edition , ,. xvii

h!troductioll to Secolld Editioll xxi

Guide to Study , xxii

1 The Economics of Insurance :............................. 1

1.1 Introduction..................................................................... 1 1.2 Utility Theory................................................................... 3

1.3 Insurance and Utility 7

1.4 Elements of Insurance......................................................... 15

1.5 Optimal Insurance 16

1.6 Notes and References.......................................................... 18

Appendix............................................................................... 19

Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 20

2 Individual Risk Models for a Short Term........................................ 27

2.1 Introduction..................................................................... 27

2.2 Models for Individual Claim Random Variables......................... 28

2.3 Sums of Independent Random Variables 34

2.4 Approximations for the Distribution of the Sum 39

2.5 Applications to Insurance.............. .. .. .. 40

2.6 Notes and References 46

Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 47

3 Survival Distributions and Life Tables............................................ 51

;", 1 Tn~rr"111rH"'n ,,1

3.2 Probability for the Age-at-Death .. .. .. .. .. 52

3.2.1 The Survival Function................................................. 52

3.2.2 Time-until-Death for a Person Age x 52

3.2.3 Curtate-future-Lifetimes.............................................. 54

3.2.4 force of Mortality.. .. . .. 55

~ ...

3.3 Life Tables 58

3.3.1 Relation of Life Table Functions to the

Survival Function ··.· ·.·· 58

3.3.2 Life Table Example 59

3.4 The Deterministic Survivorship Group..................................... 66

3.5 Other Life Table Characteristics 68

3.5.1 Characteristics , , " .. .. .. 68

3.5.2 Recursion Formulas. 73

3.6 Assumptions for Fractional Ages............................................ 74

3.7 Some AnalytIcal Laws of MortalIty 77

3.8 Select and Ultimate Tables , .. .. .. .. . . . .. .. .. . . . .. .. .. .. .. .. 79

3.9 Notes and References ·· 83

Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 84


4 ~ife Insurance 93

4.1 Introduction , 93

4.2 Illsurallces Payable at the Moment of Death.............................. 94

4.2.1 Level Benefit Insurance 94

4.2.2 Endowment Insurance ···.············ 101

4.2.3 Deferred Insurance ············ .. ··········· 103

4.2.4 Varying Benefit Insurance ·.··· .. ···· .. 105

4.3 Insurallces Payable at the End of the Year of Death 108

4.4 Relationships between Insurances Payable at the

Moment of Death and the End of the Year of Death 119

4.5 Differential Equations for Insurances Payable at the

Moment of Death , ., , , , 125

4.6 Notes and References ····· .. ·············· 126

Exercises 126

5 Life Annuities 133

. ,,.,,.,

5.1 Tnrroauctlon ······.··.················· ~JJ

5.2 Continuous Life Annuities · ··· ·· .. · 134

5.3 Discrete Life Annuities ··.··· ········ .. ··· 143

5.4 Life Annuities with m-thly Payments 149

5.5 Apportionable Annuities-Due and Complete

Annuities-Immediate 154

5.6 Notes and References ·························· 157

Exercises , ., , , , , .. ' 158

6 Benefit Prerruums ~~M

6.1 Introduction ···.··.··············· 167

6.2 Fully Continuous Premiums 170

6.2 ~._l!J D: '~Dr~~ ' .. ~~ 180

6.4 True m-thly Payment Premiums ·· .. · .. · .. ····· .. ··· 188

6.5 Apportionable Premiums 191

6.6 Accumulation-Type Benefits ······· 194

6.7 Notes and References ···· .. ······ .. ··· .. · 197

Exercises 197


7 Benefit Reserves 203

7.1 Introduction 203

7.2 Fully Continuous Benefit Reserves 206

7.3 Other Formulas for Fully Continuous Benefit Reserves 212

7.4 Fully Discrete Benefit Reserves 215

7.5 Benefit Reserves on a Semicontinuous Basis 221

7.6 Benefit Reserves Based on True m-thly Benefit

Premiums 221

7.7 Benefit Reserves on an Apportionable or Discounted

Continuous Basis 224

7.8 Notes and References 225

Exercises 225

8 Analysis of Benefit Reserves , 229 .

. 8.1 Introduction 229

8.2 Benefit Reserves for General Insurances 230

8 3 Recnrsion Relations for Fully Discrete Benefit Reserves 233

8.4 Benefit Reserves at Fractional Durations 238

8.5 Allocation of the Risk to Insurance Years , ., 241

8.6 Differential Equations for Fully Continuous Benefit

Reserves 248

8 7 Notes and References 250

Exercises " 250

9 Multiple Life Functions 257

9.1 Introduction 257

9.2 Joint Distributions of Future Lifetimes 258

9.3 The Joint-Life Status 263

9.4 The Last-Survivor Status 268

9.5 More Probabilities and Expectations 271

9.6 Dependent Lifetime Models 274

9.6.1 Common Shock 274

9.6.2 Copulas 277

9.7 Insurance and Annuity Benefits 279

9.7.1 Survival Statuses 279

9.7.2 Special Two-Life Annuities 284

9.7.3 Reversionary Annuities 285

9.8 Evaluation-Special Mortality Assumptions 287

9.8.1 Gompertz and Makeham Laws 287

9.8.2 Uniform Distribution 288

q q ~in1nlp rcmtinc""mt Prmr+iori« 291

T a

9.10 Evaluation-Simple Contingent Functions 295


~.11 NOtes ana xererences -z:77

Exercises 298

10 Multiple Decrement Models 307

10.1 Introduction 307

10.2 Two Random Variables 308

10.3 Random Survivorship Group 316



10.4 Deterministic Survivorship Group 318

10.5 Associated Single Decrement Tables 319

10.5.1 Basic Relationships 320

10.5.2 Central Rates of Multiple Decrement 321

10.5.3 Constant Force Assumption for Multiple

Decrements 322

10.5.4 Uniform Distribution Assumption for

Multiple Decrements 323

10.5.5 Estimation Issues 324

10.6 COIlSlt action of a Multiple Decrement Table 327

10.7 Notes and References 332

Exercises 333

11 Applications of Multiple Decrement Theory 341

,11.1 Introduction 341

11.2 Actuarial Present Values and Their Numerical

Evaluation .


11.3 Benefit Premiums and Reserves 345

11.4 WIthdrawal BenefIt Patterns That Can Be Ignored in

Evaluating Premiums and Reserves 346

11.5 Valuation of Pension Plans 350

11.5.l Demographic Assumptions . .. 350

11.5.2 Projecting Benefit Payment and Contribution Rates 351

11.5.3 Defined-Benefit Plans 354

11.5.4 Defined-Contribution Plans 356

11.6 Disability Benefits with Individual Life Insurance 358

11.6.1 Disability Income Benefits 358

11.6.2 Waiver-of-Premium Benefits 359

. . • ~ 'lLf'I

11.0 . .:) nenelll r reIIllUII1~ dllU rveserves vVV

11.7 Notes and References 361

Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362

12 Collective Risk Models tor a Single Period .jO/

12.1 Introduction 367

12.2 The Distribution of Aggregate Claims 368

12.3 Selection of Basic Distributions 372

12.3.1 The Distribution of N 372

12.3.2 The Individual Claim Amount Distribution 377

12.4 Properties of Certain Compound Distributions , 378

12.5 Approximations to the Distribution of Aggregate Claims 385

1,)... l'..T~~~,· ~~rl D",foT'onroc ....•..•...... . 390

Appendix 391


Exercises J7-:J

13 Collective Risk Models over an Extended Period 399

13.1 Introduction , , , , ., ., " '" 399

13.2 A Discrete Time Model 401

13.3 A Continuous Time Model 406



13.4 Ruin Probabilities and the Claim Amount Distribution 409

13.5 The First Surplus below the Initial Level 415

13.6 The Maximal Aggregate Loss " 417

13.7 Notes and References 423

Appendix 425

Exercises 430

14 Applications of Risk Theory 435

14.1 IIltroductioIl 435

14.2 Claim Amount Distributions 436

14.3 Approximating the Individual Model 441

14.4 Stop-Loss Reinsurance 445

14.5 Analysis of Reinsurance Using Ruin Theory 451

14.6 Notes and References 459

Appendix 460

Exercises 461

15 Insurance Models Includmg Expenses 465

. 15.1 Introduction 465

15.2 Expense Augmented Models 466

15.2.1 Premiums and Reserves 466

15.2.2 Accounting 469

15.3 WIthdrawal BenefIts 472

15.3.1 Premiums and Reserves 472

15.3.2 Accounting '" 476

15.4 Types of Expenses 476

15.5 Algebraic Foundations of Accounting: Single

Decrement Model 481

15.6 Asset Shares 485

15.6.1 Recursion Relations 485

15 6 2 Accounting ... . .. 487

15.7 Expenses, Reserves, and General Insurances 489

15.8 Notes and References 491

Exercises 492

16 Business and Regulatory Considerations 499

16.1 Introduction 499

16.2 Cash Values 499

16.3 Insurance Options 502

16.3.1 Paid up Insurance 502

16.3.2 Extended Term 504

16.3.3 AutomatIc PremIUm Loan 506

16.4 Premiums and Economic Considerations 507

16.4.1 Natural Premiums 508

16.4.2 Fund Objective 508

16.4.3 Rate of Return Objective 509

16.4.4 Risk-Based Objectives 511

16.5 Experience Adjustments 512



16.6 Modified Reserve Methods 515

16.7 Full Preliminary Term · · .. · 519

16.8 Modified Preliminary Term 521

16.9 Nonlevel Premiums or Benefits 523

16.9.1 Valuation ··.············· 523

16.9.2 Cash Values 526

16.10 Notes and References · · 528

Exercises 529

17 Special Annuities and Insurances 535

17.1 Introduction ······.·· 535

17.2 Special Types of Annuity Benefits 535

17.3 Family Income Insurances , , , 537

17.4 Variable Products 539

17.4.1 Variable Annuity 539

17.4.2 Fully Variable Life Insurance 541

1711':2 c.; .-l Pr<,.,·.,il1TY1 \l"r,,,}'l,,, T if", Tn<::l1r::lnre 541

17.4.4 Paid-up Insurance Increments 542

... ,~


tlexf5Te TIan i-rooucts J':tJ

17.5.1 Flexible Plan Illustration 543


17.5.2 An Alternative Design 546

A rr",l ' Renefit<:: 547

17.6.1 Lump Sum Benefits 548

17.6.2 Income Benefits · .. ··········· 550

17.7 Notes and References · · 551

Exercises 551

18 Advanced Multiple Life Theory 555

18.1 Introduction ······················· 555

18.2 More General Statuses 556

,,,,., ..... c, "ih?

iU.J '-Vl1~ v.u.u~~~ .

18.4 Contingent Probabilities and Insurances 564

18.5 Compount Contingent Functions ··· ::>00

18.6 More Reversionary Annuities , " " 570

18.7 Benefit Premiums and Reserves · .. 573

18.8 Notes and References 575

Appendix .. ' , , , 577

Exercises , 578

~ ,. ~, I';QC:;


rupUlclllUll 1.lleuly ............••........................•.••..........................

19.1 Introduction ·······.············· 585

19.2 The Lexis Diagram · · .Jb::l

1Q.3 A Continuous Model 587

19.4 Stationary and Stable Populations ( 593

19.5 Actuarial Applications 595

19.6 Population Dynamics 599

19.7 Notes and References 603

Exercises 603


20 Theory of Pension Funding 607

20.1 Introduction 607

20.2 The Model 608

20.3 Terminal Funding 609

20.4 Basic Functions for Retired Lives 611

20.4.1 Actuarial Present Value of Future Benefits, (rA)t 611

20.4.2 Benefit Pay ment Rate, Bt 611

20.4.3 The Allocation Equation 612

20.5 Accrual of Actuarial Liability 614

20.6 Basic Functions for Active Lives 615

20.6.1 Actuarial Present Value of Future Benefits, (aA)t 615

20.6.2 Normal Cost Rate, P, 616

20.6.3 Actuarial Accrued Liability, (aV)t 618

20.6.4 Actuarial Present Value of Future Normal

Costs, (Pa)t 619

20.7 Individual Actuarial Cost Methods , 622

20.8 Group Actuarial Cost Methods 624

20.9 Basic Functions for Active and Retired Members

CombIned 628

20.10 Notes and References 630

Exercises 630

21 Interest as a Random Value 635

21.1 Introduction 635

21.1.1 Incorporating Variability of Interest 636

21.1.2 Notation and Preliminaries 637

21.2 Scenarios 638

21.2.1 Deterministic Scenarios 638

21 22 Random Scenarios' Deterministic Interest Rates 641

21.3 Independent Interest Rates 643

21.4 Dependent Interest Rates 649

21.4.1 Moving Average Model 649

21.4.2 Implementation " , , ., , 654

21.5 Financial Economics Models 655

21.5.1 Information in Prices and Maturities 655

21.5.2 Stochastic Models 659

21.6 Management of Interest Risk 663

21.6.1 Immunization 663

21.7 Notes and References 666

Appendix 2A Appendix 2B Appendix 3 Appendix 4

Illustrative Life Table 675

Illustrative Service Table 685

Symbol Index 687

General Rules for Symbols of Actuarial Functions 693

Appendix 6 Appendix 7

Some Mathematical Formulas Useful in Actuarial

Mathematics 701

Bibliography 705

Answers to Exercises ··.················· 719

Appendix 5

Index 749


NEWTON L. BOWERS, JR., Ph.D., F.s.A., M.A.AA, received a as. from Yale

University and a Ph.D. from the University of Minnesota. After his graduate studies, be joined the faclllty at the I Iniversity of Michigan and later moved to Drake

University. He held the position of Ellis and Nelle Levitt Professor of Actuarial

Science until his retirement in 1996.

HANS U. GERBER, PhD., AS.A, received the Ph.D. degree from the Swiss Federal Institute of Technology. He has been a member of the faculty at the Universities of Michigan and Rochester. Since 1981 he has been on the faculty at the Business School, University of Lausanne, where he is currently head of the Institute of Actuarial Science."

JAMES C. HICKMAN, Ph.D., F.s.A., A.c. A.5. , M.AA.A., received a B.A from

Simpson College and a PhD. from the University of Iowa. He was a member of the faculty at the University of Iowa and at the University of Vlisconsin, Madison,

until his retirement in 1993. For five years he was Dean of the School of Business

at ISCOnSIn.

DONALD A. JONES, PhD., F.S.A., E.A, M.A.A.A., received a B.5. from Iowa State University and M.S. and Ph.D. degrees from the University of Iowa. He was a member of the actuarial faculty at the University of Michigan and a working partner of Ann Arbor Actuaries until 1990 when he became Director of the Actuarial Science Program at Oregon State University.

CECIL J NESBITT Ph D . F S.A . M A A A, received his mathematical education

at the University of Toronto and the Institute for Advanced Study in Princeton. He taught actuarial mathematics at the Ulliversity of Michigall fwm 1938 to 1980. He

served the Society of Actuaries from 1985 to 1987 as Vice-President for Research and Studies."

"Professors Gerber and Nesbitt were involved as consultants with the revisions incorporated in the second edition.

Authors' Bi



Introduction to First Edition*

This text IepIesellts a first step in communicating the revolution in the actuarial

profession that is taking place in this age of high-speed computers. During the

short period of time since the inventIon of the mIcrochIP, actuaries have been fIeed from numerous constraints of primitive computing devices in designing and man-

aging insurance systems. They are now able to focus more of their attention on creative solutions to society's demands for financial security.

To provide an educational basis for this focus, the major objectives of this work are to integrate life contingencies into a full risk theory framework and to demonstrate the wide variety of constructs that are then possible to build from basic models at the foundation of actuarial science A ctuaria1 science is ever evolving,

and the procedures for model building in risk theory are at its forefront. Therefore,

we examine the nature of models before proceeding with a mOle detailed discus f the text.

Intellectual and physical models are constructed either to organize observations into a comprehensive and coherent theory or to enable us to simulate, in a laboratory or a computer system, the operation of the corresponding full-scale entity. Models are absolutely essential in science, engineering, and the management of large organizations. One must, however, always keep in mind the sharp distinction betvv'een a model and the reality it reprE2sents A satisfactory model captures enough

of reality to give insights into the successful operation of the system it represents.

ened our insights about insurance systems.

*Chapter references and nomenclature have been changed to be in accord with the second edition. These changes are indicated by italics.


before us the idea that real insurance systems operate in an environment that is more complex and dynamic than the models studied here. Because models are only approximations of reality, the work of model building is never done; approximations can be improved and reality may shift. It is a continuing endeavor of any scientific discipline to revise and update its basic models. Actuarial science is no exception.

Actuarial science developed at a time when mathematical tools (probability and

calculus, in particular), the necessary data (especially mortality data m the form of life tables), and the socially perceived need (to protect families and businesses from

the financial consequences of untimely death) coexisted. The models constructed at the genesis of actuarial science are still useful. However, the general environment in which actuarial science exists continues to change, and it is necessary to periodically restate the fundamentals of actuarial science in response to these changes.

We illustrate this with three examples:

1. The insurance needs of modern societies are evolving, and, in response, new

systems of employee benefits and social insurance have developed. New models for tlrese systems have been needed alLd corrstrtrcted:

2. Mathematics has also evolved, and some concepts that were not available for

use in building the original foundations of actuarial science are now part of a general mathematics education. If actuarial science is to remain in the main-

stream of the applied sciences, it is necessary to recast basic models in the language of contemporary mathematics.

3. Finally, as previously stated, the development of high-speed computing equipment has greatly increased the ability to manipulate complex models. This has far-reaching consequences for the degree of completeness that can be incorporated into actuarial models

'"T't • _L _ 1 .1 C -' d. • r

~L"u 'VU' H_ULUH_u uLVUCl'" ULaL ci i c lUlLuallLClLLal LV LlLC LUllClLL t-'laLLlLC vi ctLLU-

arial science. They are explored with tools acquired in the study of mathematics,

in particular, undergraduate level calculus and probability. The proposition guiding Chapters 1 14 is that there is a set of basic models at the heart of actuarial

science that should be studied by all students aspiring to practice within any of the various actuarial specialities. These models are constructed using only a limited number of ideas. We will find many relationships among those models that lead to a unity in the foundations of actuarial science. These basic models are followed, in Chapters 15-21, by some more elaborate models particularly appropriate to life insurance and pensions.

While this book is intended to be comprehensive, it is not meant to be exhaustive.

in oruer to avoic any mlsunderstandmg, we Wlll truncate me hmitations or the text: • Mathematical ideas that could unify and, in some cases, simplify the ideas

presented, but which are not included in typical undergraduate courses, are not used. For example, moment generating functions, but not characteristic functions, are used in developments regarding probability distributions. Stieltjes integrals, which could be used in some cases to unify the presentation


Introduction to First Edition

of discrete and continuous cases, are not used because of this basic decision on mathematical prerequisites.

• The chapters devoted to life insurance stress the randomness of the time at which a claim payment must be made. In the same chapters, the interest rates used to convert future payments to a present value are considered deterministic and are usually taken as constants. In view of the high volatility possible in ilLterest rates, it is Hatural to ask why probability models for interest rates '.vere

not incorporated. Our answer is that the mathematics of life contingencies on

a probabilistic foundation (except for interest) does not mvolve Ideas beyond those covered in an undergraduate program. On the other hand, the modeling

of interest rates requires ideas from economics and statistics that are not included in the prerequisites of this volume. In addition, there are some technical problems in building models to combine random interest and random time of claim that are in the process of being solved.

• Methods for estimating the parameters of basic actuarial models from observations are not covered. For example, the construction of life tables is not

• This is not a text on computing. The issues involved in optimizing the orga-

nization of input data and computation In actuanal models are not dIscussed. This is a rapidly changing area, seemingly best left for readers to resolve as

they choose in light of their own resources.

• MallY important actuarial problems created by long term practice and insur-

ance regulation are not discussed. This is true in sections treating topics such as premiums actually charged for life insurance policies, costs reported for pensions, restrictions on benefit provisions, and financial reporting as required by regulators.

• Ideas that lead to interesting puzzles, but which do not appear in basic actuarial models, are avoided. Average age at death problems for a stationary population

do not appear for thIs reason.

This text has a number of features that distinguish it from previous fine textbooks

on life contingencies. A number of these features represent decisions by the authors on material to be included and will be discussed under headings suggestive of the

topics involved.

Probability Approach

As indicated earlier, the sharpest break between the approach taken here and that taken in earlier English language textbooks on actllarial mathematics is the

much fuller use of a probabilistic approach in the treatment of the mathematics of hfe contmgenCIes. Actuanes have usually written and spoken of applyilLg proba-

bilities in their models, but their results could be, and often were, obtained by a

deterministic rate approach. In this work, the treatment of life contingencies is based on the assumption that time-until-death is a continuous-type random variable. This admits a rich field of random variable concepts such as distribution function, probability density function, expected value, variance, and moment generating function. This approach is timely, based on the availability of high-speed


computers, and is called for, based on the observation that the economic role of life insurance and pensions can be best seen when the random value of time-untildeath is stressed. Also, these probability ideas are now part of general education in mathematics, and a fuller realization thereof relates life contingencies to other fields of applied probability, for example, reliability theory in engineering.

Additiorratly; the determillistic rate approach is described for completelless aud

is a tool in some developments. However, the results obtained from using a deter-

ministic model usually can be obtained as expected values in a probabilistic model.

Integration with Risk Theory

Risk theory is defined as the study of deviations of financial results from those expected and methods of avoiding inconvenient consequences from such deviations. The probabilistic approach to life contingencies makes it easy to incorporate long-term contracts into risk theory models and, in fact, makes life contingencies

only a part, but a very important one, of risk theory. Ruin theory, another important part of risk theory, is included as it pro" ides insight into one source, the insurance

claims, of adverse long-term financial deviations. This source is the most unique

aspect of models for insurance enterprises.

Utility Theory

This text contains topics on the economics of insurance. The goal is to provide a motivation, based on a normative theory of individual behavior in the face of uncertainty, for the study of insurance models. Although the models used are highly simplified, they lead to inSights into the economic role of insurance, and to an

appreciation of some of the issues that arise in making insurance decisions.

Consistent Assumptions

The assumption of a uniform distribution of deaths in each year of age is consistently used to evaluate actuarial functions at nonintegral ages. This eliminates some of the anomalies that have been observed when inconsistent assumptions are applied in situations involving high interest rates.

rvewton L. oo vve rs Hans U. Gerber

James C. Hickman Donald A. Tones

Cecil J. Nesbitt

tntrridt ir tirvn rr, i=ir<:t Friitinn


Introduction to Second Edition

Actuarial science is not static. In the time since the publication of the first edition of Actuarial Mathematics, actuarial science has absorbed additional ideas from economics and the mathematical sciences. At the same time, computing and communications have become cheaper and faster, and this has helped to make

feasible more complex actuarial models. During this period the financial risks that modern societies seek to manage have also altered as a result of the globalization

of business, technological advances, and political shifts that have changed public

po icies.

It would be impossible to capture the full effect of all these changes in the revision of a basic textbook. Our objective is more modest, but we hope that it is realistic. This edition is a step in an ongoing process of adaptation designed to keep the fundamentals of actuarial science current with changing realities.

In the second edition, changes in notation and nomenclature appear in almost every section. There are also basic changes from the first edition that sholiid be


1. CommutatIOn functIons, a classIc tool In actuarial calculations, are not used.

This is in response to the declining advantages of these functions in an age

when interest rates are often viewed as random variables, or as varying deterministically, and the probability distribution of time until decrement may depend on variables other than attained age. Starting in Chapter 3, exercises that illustrate actuarial calculations using recursion formulas that can be implemented with current software are introduced. It is logically necessary that the challenge of implementing tomorrow's software is left to the reader.

2 Utility theory is no longer confined to the first chapter Examples are given that

illustrate how utility theory can be employed to construct consistent models

for premiums and reserves that differ from the conventional model that implicitly depends on linear utility of wealth.

3. In the first edition readers were seldom asked to consider more than the first and second moments of loss random variables. In this edition, following the intellectual path used earlier in physics and statistics, the distribution functions and probability density functions of loss variables are illustrated.

4. The basic material on reserves is now presented in two chapters. This facilitates a more complete development of the theory of reserves for general life insurances with varying premiums and benefits

5. In recent years considerable actuarial research has been done on joint distributions for several future lifetime ralldom variables where mutual indeperr-

dence is not assumed. This work influences the chapters on multiple life ac-

tuarial functions and multiple decrement theory.

6. There are potentially serious estimation and interpretation problems in multiple decrement theory when the random times until decrement for competing causes of decrement are not independent. Those problems are illustrated in the second edition.

Authors' Introductions and Guide to Stud


7. The applications of multiple decrement theory have been consolidated. No attempt is made to illustrate in this basic textbook the variations in benefit formulas driven by rapid changes in pension practice and regulation.

S. The confluence of new research and computing capabilities has increased the use of recursive formulas in calculating the distribution of total losses derived from risk theory models. This development has influenced Chapter 12.

9. Ihe matenal on pncmg hfe msurance wIth death and wIthdrawal benefits and accounting for life insurance operations has been reorganized. Business and

regulatory considerations have been concentrated in one chapter, and the foundations of accounting and provisions for expenses in an earlier chapter. The

discussion of regulation has been limited to general issues and options for addressing these issues. No attempt has been made to present a definitive interpretation of regulation for any nation, province, or state.

10. The models for some insurance products that are no longer important in the market have been deleted. Models for new products, such as accelerated benefits for terminal illness or long-term care, are introduced.

11. The final chapter contains a brief introduction to simple models in which in-

terest rates are random variables. In addition, ideas for managing interest rate

risk are discussed, It is hoped that this chapter will provide a bridge to recent developments within the intersection of actllarial mathematics and financial


As the project of writing this second edition ends, it is clear that a significant new development is under way. This new endeavor is centered on the creation of general models for managing the risks to individuals and organizations created by uncertain future cash flows when the uncertainty derives from any source. This blending of the actuarial/statistical approach to building models for financial security systems with the approach taken in financial economics is a worthy assign-

ment for the next cohort of actuarial students.

• T n

!"\JeWlUIl L. UUV,Cl::>

James C. Hickman Donald A. Jones

Guide to Study

The reader can consider this text as covering the two branches of risk theory.

Individual risk theory views each policy as a unit and allows construction of a

rY"O~.-lal few '> .,. .. ,.., .. ~,..,f ~~l;r;ac h" ~.-l.-l;~.,. !-ha f;~'>~r;~l .. ac"l~c f~ .. !-ha ~ ·-,~a ,,,Iir;p,,

o 't r - ] ·0 '? r

in the group. Collective risk theory uses a probabilistic model for total claims that

avoids the step of adding the results for individual policies. This distinction is sometimes difficult to maintain in practice. The chapters, however, can be classified as illustrated below.

Individual Risk Theory

Collective Risk Theory

1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 15, 16, 17, 18, 21

12, 13, 14, 19,20

It is also possible to divide insurance models into those appropriate for shortterm insmance. where investment income is not a significant factor, and long-term

insurance, where investment income is important. The following classification scheme provides this division of chapters along "'lith an additional division of long-

term models between those for life insurance and those for pensions.

Long-Term Insurances

Short-Term Insurances

Life Insurance


1, 2, 12, 13, 14

3, 4, 5, 6, 7, 8, 9, 10, 11 15 16 17 18,21

9, 10, 11, 19, 20, 21

The selection of topics and their organization do not follow a traditional pattern.

As stated previously, the new organization arose from the goal to first cover ma-

terial considered basic for all actuarial students (Chapters 1-14) and then to include a more in-depth treatment of selected topics for students specializing in life insurance and pensions (Chapters 15-21).

The discussion in Chapter 1 is devoted to two ideas: that random events can disrupt the plans of decision makers and that insurance systems are designed to reduce the adverse fmanCIal effects of these events. To illustrate the latter, sillgle

insurance policies are discussed and convenient, if not necessarily realistic, distri-

butions of the loss random variable are used. In subsequent chapters, more detailed models are constructed for use with insurance systems

In Chapter 2, the individual risk model is developed, first in regard to single policies, then in regard to a portfolio of policies. In this model, a random variable, S, the total claims in a single period, is the sum of a fixed number of independent random variables, each of which is associated with a single policy. Each component of the sum S can take either the value 0 or a random claim amount in the course

From the viewpoint of risk theory, the ideas developed in Chapters 3 through 11 can be seen as extending the ideas of Chapter 2. Instead of considering the

potential claims in a short period from an individual policy, we consider loss variables that take into account the financial results of several periods. Since such random variables are no longer restricted to a short time period, they reflect the time value of money. For groups of individuals, we can then proceed, as in

I i



Authors' Introductions and Guide to Study

Chapter 2, to use an approximation, such as the normal approximation, to make probability statements about the sum of the random variables that are associated with the individual members.

In Chapter 3, time-of-death is treated as a continuous random variable, and, after defining the probability density function, several features of the probability

dlstnbuhon are mtroduced and explored. In Chapters 4 and 5, hfe Insurances and annuities are introduced, and the present values of the benefits are expressed as

functions of the time-of-death. Several characteristics of the distributions of the present value of future benefits are eXamined. In Chapter 6, the equivalence prin-

ciple is introduced and used to define and evaluate periodic benefit premiums. In Chapters 7 and 8, the prospective future loss on a contract already in force is investigated. The distribution of future loss is examined, and the benefit reserve is defined as the expected value of this loss. In Chapter 9, annuity and insurance contracts involving two lives are studied. (Discussion of more advanced multiple life theory is deferred until Chapter IS.) The discussion in Chapters 10 and 11

investigates a more realistic model in which several causes of decrement are pos sible. In Chapter 10, basic theory is examined, whereas in Chapter 11 the theory is

applied to calculating actuarial present values for a variety of insurance and pen-

In Chapter 12, the collective risk model is developed with respect to single-period

considerations of a portfolio of policies. The distribution of total claims for the period is developed by postulating the characteristics of the portfolio in the aggregate rather than as a sum of individual policies. In Chapter 13, these ideas are extended to a continuous-time model that can be used to study solvency requirements over a long time period. Applications of risk theory to insurance models are given an overview in Chapter 14.

Elaboration of the indi"idllal model to incorporate operational constraints sllch

as acquisition and administrative expenses, accounting requirements, and the effects of contract terminations is treated in Chapters 15 and 16. In Chapter 17, in-

dividual risk theory models are used to obtain actuarial present values, benefit and contract premiums, and benefit reserves for selected special plans including life annuities with certain periods that depend on the contract premium, variable and flexible products, and accelerated benefits. In Chapter 18, the elementary models for plans involving two lives are extended to incorporate contingencies based on a larger number of lives and more complicated benefits.

then applied to tracing the progress of life insurance benefits provided on a group,

progress of retirement income benefits provided on a group basis in Chapter 20.

Chapter 21 is a step into the future. Interest rates are assumed to be random variables. Several stochastic models are introduced and then integrated into models for basic insurance and annuity contracts.

The following diagram illustrates the prerequisite structure of the chapters. The arrows indicate the direction of the flow. For any chapter, the chapters that are upstream are prerequisite. For example, Chapter 6 has as prerequisites Chapters 1,

2, 3, 4, and 5. .~.

t ~----~0 t



We have a couple of ltilLt5 for the reader, particularly for one for whom the

material is new. The exercises are an important part of the text and include material not covered in the main discussion. In some cases, hints will be offered to aid in the solution. Answers to all exercises are provided except where the answer is given in the formulation of the problem. Writing computer programs and using electronic spreadsheets or mathematical software for the evaluation of basic formulas are excellent ways of enhancing the level of understanding of the material. The student

IS encouraged to use these tools to work through the computing exercises.

We conclude these introductory comments with some miscellaneous information on the format of the text. First, each chapter concludes with a reference section that

provides guidance to those who wish to pursue further study of the topics covered in the chapter. These sections also contain comments that relate the ideas used in insurance models to those used in other areas.

Second, Chapters 1, 12, 13, 14, and 18 contain some theorems with their proofs included as chapter appendices. These proofs are included for completeness, but

Authnr<:' Intrnr/"rtinn<: :::anrl (" lirlo tn ,t,;;/;/


are not essential to an understanding of the material. They may be excluded from study at the reader's discretion. Exercises associated with these appendices should also be considered optional.

Third, general appendices appear at the end of the text. Included here are numerical tables for computations for examples and exercises, an index to notation,

a discussion of general rules for writing actuarial symbols, reference citations, allswers to exercises, a subject index, and suppl€ID€ntal math€matical formulas that

are not assumed to be a part of the mathematical prerequisites.

Fourth, we observe two notational conventions. A referenced random variable, X, for example, is designated with a capital letter. This notational convention is not used in older texts on probability theory. It will be our practice, in order to indicate the correspondence, to use the appropriate random variable symbol as a subscript on functions and operators that depend on the random variable. We will use the general abbreviation log to refer to natural (base e) logarithms, because a distinction

between natural and common logarithms is unnecessary in the examples and exercises. We assume the natural logarithm in our computations

Fifth, currencies such as dollar, pound, lira, or yen are not specified iII the examples and exercises due to the international character of the required


Finally, since we have discussed prerequisites to this work, some major theorems from undergraduate calculus and probability theory will be used without review or restatement in the discussions and exercises.



1.1 Introduction

follow. However, experience teaches that plans will not unfold with certainty and

Each of us makes plalls alld has expectations about the path his or her life will

sometimes expectations will not be reahzed. OccasIOnally plans are frustrated because they are built on unrealistic assumptions. In other situations, fQ!hdtou5 cir-

cumstances interfere.Jnsurance is designed to protect against serious financiaLre't~_r~al~Jhatresult fromnmdom event~ intruding on the plans of individuals.

We should understand certain basic limitations on insurance protection. First, iL is restricted to reducing those consequences of random events thatcan be measured in monetary terms. Other types of losses may be important, but not amenable to reduction through insurance.

For example, pam and suffenng may be caused by a random event. However, insurance coverages designed to compensate for pain and suffering often have been

troubled by the difficulty of measuring the loss in monetary units. On the other hand, economic losses can be caused by events such as property set on fire by its

owner. Whereas the monetary terms of such losses may be easy to define, the events .are not insurable because of thenonrandom nature of creating the losses.

A second basic limitation is that insurance doesnot directly reduce the probability _ of loss. The existence of windstorm insurance will not alter the probability of a destructive storm. However, a well-designed insurance system often provides

fmanClal mcenflves for loss prevention activities. An insurance product that ellcouraged the destruction of property or the withdrawal of a productive person

from the labor force would affect the probability of these economically adverse events Such insurance would not be in the public interest.

Several examples of situations where random events may cause financial losses are the following:

• The destruction of property by fire or storm is usually considered a random event in which the loss can be measured in monetary terms.

Chapter 1 The Economics of Insurance


• A damage award imposed by a court as a result of a negligent act is often considered a random event with resulting monetary loss .

• Prolonged illness may strike at an unexpected time and result in financial losses. These losses will be due to ~;<tra health care expens£s and reduced earned income.

• The death of a young adult may occur while long-term commitments to family

or business remain unfulfilled. Or, if the individual survives to an advanced age, resources for meeting the costs of living may be depleted.

These examples are designed to illustrate the definition:

An insurance system is a mechanism for reducing the adverse financial impact of random events that prevent the fulfillment of reasonable expectations.

It is helpful to make certain distinctions between insurance and related systems.

Banking institutions were developed for the purpose of receiving, investing, and dispensing the savings of individuals and corporations. The cash flows in and out 9fa §avings iIlstitution do not follow deterministic paths. However, unlike insur-

ance systems, savings institutions do not make payments based on the size of a

financial loss occurring from an event outside the control of the person suffering

Another system that does make payments based on the occurrence of random

events is gambling. Gambling or wagering, however, stands in contrast to an insurance system in that an insurance system is designed to protect against the economic impact of risks that exist independently of, and are largely beyond the control of, the insured. The typical gambling arrangement is established by defining payoff rules about the occurrence of a contrived event, and the risk is voluntarily sought by the participants. Like insurance, a gambling arrangement typically re-

distributes wealth, but it is there that the similarity ends.

Our definition of an insurance system is purposefully broad. It encompasses systems that COvel losses ill both property alld htrrrran=life values. It is illtellded to

cover insurance systems based on individual decisions to participate as well as systems where participation is a condition of employment or residence. These ideas are discussed in Section 1.4.

Ih~~~5ln()!TIic justification for an insurance system ~ that_itcontributest() geI1~~~1 welfare by improving the prospect that_plans will not be frustrated by random .

. events. Suc~ systems may also _increase total production by encouraging individuals and corporations to embark on ventures where the possibility of large losses would

inhibit such projects in the absence of insurance. The development of marine insurance, for reducing the, financial impact of the perils of the sea, is an example of

this point. Foreign trade permitted specialization and more efficient production, yet mutually advantageous trading activity might be too hazardous for some potential trading partners without an insurance system to cover possible losses at se~.


Section 1.1 Introduction

1.2 Utility Theory

If people could foretell the consequences of their decisions, theJt lives would be simpler but less interesting. We would all make decisions on th~-ba~is~(p;-eferences for certain consequences. However, we do not possessperfect fog~_sighLAt_ best, we can select an action that will lead toone '~~-t of uncertainties rather than

ano·th~r. An el~bQrate_theory has.been -d.evelop~dthat providesinsightsjnto deci.§iQI) ffi,!_l<i_Il$. in _!he _face of. uncertainty -, This. bQdy:.Of. knmv:.ledg€l_ is c.alledJltility


One solution to the problem of decision making in the face of uncertainty is to define the value of an economic project with a random outcome to be its expected value. ~_this expected _VfllJ!:eprinCiPI~lJhe_distrihutionoLpQs_~iQ!e .9_l-l_t<:.0_Il!.es..J2l_'!y be replaced for decision purposes bya single number, the expected value of the rcii1doU; ~onetary outcomes. By th.js _ _PXLD.cipJe!_w~> dec!si9!:l~ ITl,!k~r V\l'Q\llqJ?_~~jndif-

krent between.assuming.the.racdcm.css X ang_p-_~yiDg2mQ_llI1t E[XliD_prder ~2_ b~ _re_lieved of the possible ~ Similarl)~ a decision maker would be willing to pay

up to E[Y] to participate in a gamble with random payoff Y. In economics @l~ i~~g_e4_yaJ_1:l~ ot r~I:t(19iri prospects. with· monetary paymerit$ is frequeIltly called

the fair or actuarial value of the prospect.

Many decision makers do not adopt the expected value principle. For them, their wealth level and other aspects of the distribution of outcomes influence their decisions.

Below is an illustration designed to show the inadequacy of the expected value principle for a decision maker considering the value of accident insurance In all

cases, it is assumed that the probability of an accident is 0.01 and the probability of no accident is 0.99. Three cases are considered accordiIlg to the amouIIt of loss

arising from an accident; the expected loss is tabulated for each.


Possible Losses

Expected Loss

1 2 3

o 1

o 1000

o 1000000

0.01 10.00 10000.00

A loss of 1 might be of little concern to the decision maker who then might be llnwilling to pay more than the expected loss to obtain insurance However the

loss of 1,000,000, which may exceed his net worth, could be catastrophic. In this case, the decisioIl maker might well be willillg !Q pay llLore thaI! ~lte expected loss

of 10,000 in order to obtain insurance. The fad that the amountadecision maker would pay for protection against a random 10~s';1n;y differ from the eX2ec_~~d~~1~e ; suggests that the expected value principle is iriadequate to model behavior.-------c~

Chapter 1 The Economics of Insurance


We now study another approach to explain why a decision maker may be willing t~ _ _p_ay more than the expected value. bt first we simply assume thatt~~ly~20r ~!~~hat .. c:l_partic.llJ~rdedsion maker attaches to wealth of amoun(~measured in monetary units, can be specified in the form of a function u(w), called a utility function. We demonstrate a procedure by which )i-fewvalues- of such a function can be determined. For this we assume that our decisio;-maker has wealth equal

to 20,000. A linear transformatIon,

7?? yields ffiriiti6n:-u*(,i1)~whjai [5 -essentiany~equTvalerlLfOu(w), It then follows by


ti u(w) + b

a> 0,

choice of a and b that we can determine arbitrarily the 0 point and one additional point of an individual's utility function. Therefore, we fix u(O) = -1 and u(20,OOO) = O. These values are plotted on the solid line in Figure 1.2.1.

Determination of a Utility Function



wea i thousands

We now ask a question of our deClslOn maker: Suppose you face a loss of 20,000 with probability 0.5, and will remain at your current level of wealth with proba-

bility 0.5. What is the maximum amount" G you would be willing to pay for

'Premium quantities, by convention in insurance literature, are capitalized although they are not random variables.

complete insurance protection against this random loss? We can express this question in the following way: For what value of G does

u(20,OOO - G) = 0.5 u(20,OOO) + 0.5 u(O)

= (0.5)(0) + (0.5)( -1) = -0.5?

If he pays amount G, his wealth will certainly remain at 20,000 - G. The equal

sign mdicates that the deCISIOn maker is mdifferent between paying G with certainty and accepting the expected utility of wealth expressed on the right-hand


Suppose the decision maker's answer is G = 12,000. Therefore,

u(20,OOO - 12,000) = u(8,OOO) = -0.5.

This result is plotted on the dashed line in Figure 1.2.1. Perhaps the most important aspect of the decision maker's response is that he is willing to pay an amount for insurance that is greater than

(0.5)(0) + (0.5)(20,000) = 10,000,

the expected value of the loss.

This procedure can be used to add as many points [w, u(w)], for 0 :5 W :5 20,000, as needed to obtain a satisfactory approximation to the decision maker's utility of

wealth function. Once a utility value has been assigned to wealth levels WI and w2, where 0 :5 WI < W2 ::::: 20,000, we can determine an additional point by asking the decision maker the following question: What is the maximum amount you would pay for complete insurance against a situation that could leave you with wealth W2 with specified probability p, or at reduced wealth level WI with probability 1 - p? We are asking the decision maker to fix a value G such that

(1.2.1 )

Once the value W2 - G - W3 is available, the point [W3' (1 - p)u(wI) + p u(w2)] is determined as another point of the utility function. Such a process has been used

to assign a fourth point (12,500, -0.25) in Figure 1.2.1. Such solicitation of preferences leads to a set of points on the decision maker's utility function. A smooth function with a second derivative may be fitted to these points to provide for a utility function everywhere.

After a decision maker has determined yhis utility of wealth function by_th~_ nietltod ou-tlined,_th_e fUflc:tion can be used to compare t'vyo rand()ffi eC_QDornl£prps

pects. The prospects are denoted by the random variables X and Y. ;!!__e_ se~!a

deCiSIOn rule _that IS consistent WIth the preferenc;:es already elicited in the .determfilatiO~- ~Tthe utility of wealth function, Thus, if the decision maker has wealth

w, and must compare the random prospects X and y, the decision maker selects X if

E[u(w + X)] > E[u(w + Y)], and the decision maker is indifferent between X and Y if


E[u(w + X)] = E[u(w + Y)].

Although the method of eliciting and using a utility function may seem plausible, it is clear that our informal development must be augmented by a more rigorous chain of reasoning if utility theory is to provide a coherent and comprehensive framework for decision making in the face of uncertainty. If we are to understand the economic role of insurance, such a framework is needed. An outline of this

more rigorous theory follows.

The theory starts with the assUIr~p~ion that a rational decision maker, when faced

with two distributions of outcomes affecting wealth, is able to express a preference for o~~ of the distributions or indifference between them. Furthermore, th~pre{ erences must satisfy tcertain consistency requirements) Jh~ _ _!l1e_ory _ culminates in a theorem stating that _if preferences satisfy the consistency requirements, _ther~~ a 1:!!m~y_function t1wJsuch that if the distribution of X is preferred to thedtstribution of Y, E[u(X)] > f[u(Y)I and if the decision maker is indifferent between.the _tw,o distributions. E[u(X)] = E[u(Y)]. That is, the qualitative preference or indifference

relation may be replaced by a consistent numerical comparison. In SectIon 1.6,


1. Utility theory is built on the assumed existence and consistency of preferences for probability distributions of outcomes. A utility function should reveal no surprises. It is a numerical description of existing preferences.

u*(w) = a u(w) + b

a> 0,


E[u(X)] > E[u(Y)]

is equivalent to

E[u*(X)] > E[u*(Y)].

function is an increasing linear

u(w)=aw+b a>O.

Then, if E[X] = fLx and E[y] = fLy, we have

E[u(X)] = a fLx + b > E[u(Y)] = a fLy + b

if and only if fLx > fLy. That is, for increasing linear utility functions, preferences

for istributions of outcomes are in t e same order as the expected values of the distributions being compared. Therefore, the expected value principle for ra_ tional economic behavior in the face of uncertainty i~ consistent with the expected utility rule when the utility function is an increasing linear one.

1.3 Insurance and Utilit

the economic role of insurance. To examine this role we start with an illustration.

uppose a ecision ma er owns a proper y at may e amage or estroye In the next accounting period. The amount of the loss, which may be 0, is a random variable denoted by X. We assume that the distribution of X is known. Then E[X], the expected loss in the next period, may be interpreted as the long-term average loss if the experiment of exposing the property to damage may be observed under identical conditions a great many times. It is clear that this long-term set of trials could not be performed by an individual decision maker.

the financial consequences of the damage or destruction of property. The insurer

s ctes a wou promise 0 pay e owner 0 a proper

a defined amount equal to or less than the financial loss if the property were dam-

aged or destroyed during the period of the policy. The contingent payment linked to the amount of the loss is called a claim payment. In return for the promise contained in the policy, the owner of the property (insured) pays a consideration (premium).

The amount of the premium payment is determined after an economic decision principle bas been adopted by each of the imurer and insured. An opportunity

exists for a mutually advantageous insurance policy when the premium for the

policy set by the insurer is less than the maximum amount that the property owner is willing to pay for insurance.

Within the range of financial outcomes for an individual insurance policy, the insurer's utility function might be approximated by a straight line. In this case, the insurer would adopt the expected value principle in setting its premium, as indicated in Section 1.2, Observation 3; that is, the insurer would set its basic price for full insurance coverage as the expected loss, E[X] = p., In this context fJ.. is called the pure or net premium for the I-period insurance policy To provide for expenses,

taxes, and profit and for some security against adverse loss experience, the insurance system would decide to set the premium for the policy by loading, adding

to, the pure premium. For instance, the loaded premium, denoted by H, might be

gIven y

H = (1 + a)fJ.. + C

a > 0,

C> O.

In this expression the quantity afJ.. can be viewed as being associated with expenses that vary with expected losses and with the risk that claims experience will deviate from expected. The constant c provides for expected expenses that do not vary with

Chapter 1 The Economics of Insurance


losses. Later, we will illustrate other economic principles for determining premiums that might be adopted by the insurer.

We now apply utility theory to the decision problems faced by the owner of the property subject to loss. Ihe property owner has a utility of wealth function u(w)

where wealth 7IJ is measured in monetary terms. The owner faces a possible Loss

due to random events that may damage the propel ty. The distribution of the random loss X is assumed to be known. Much as in (1.2.1), the owner will be indif-

ferent between paying an amount G to the insurer, who wIll assume the random financial loss, and assuming the risk himself. This situation can be stated as

The right-hand side of (1.3.1) represents the expected utility of not buying insurance when the owner's current wealth is w. The left-hand side of (1.3.1) represents the

expected utihty of paying G for complete financial protection.

u(w - G) = E[u(w - X)].

(1.3.1 )

If the owner has an increasing hnear utility function, that is. u(w) bw + d with

b > 0, the ovmer wiJl be adopting the expected value principle. In this case the

owner prefers, or is indifferent to, the insurance when

b(w G) + d > E[u(w X)]

b(w - G) + d"2 b(w - u.) + d,

G s u..

That is, if the owner has an increasing linear utility function, the premium payments that will make the owner prefer, or be indifferent to, complete insurance are less than or equal to the expected Io!;!> In the absence of a subsidy, an insurer, over

the long term, must charge more than its expected losses. Therefore, in this case,

there seems to be little opportumty for a mutually advantageous insurance contract. If an insurance contract is to result, the insurer must charge a premium in excess

of expected losses and expenses to avoid a bias toward insufficient income The property owner then cannot use a linear utility function.

In Section 1.2 we mention that the preferences of a decision maker must satisfy certain consistency requirements to ensure the existence of a utility function. Although these requirements were not listed, they do not include any specifications that wouLd force a utility function to be linear, quadratic, exponential, logarithmic,

or any other particular form. In fact, each of these named functIOnS might serve as

a utility function for some decision maker or they might be spliced together to leflect some other decision maker's preferences.

Nevertheless, it seems natural to assume that u(w) is an increasing function, "more is better." In addition, it has been observed that for many decision makers, each additional equal increment of wealth results in a smaller increment of associated utility. This is the idea of decreasing marginal utility in economics.

u(w G)


X) + d],

with positive slopes. It is such that ~2!1(W) <:: o. It these ideas are extended to

the approximate utility function of Figure 1.2.1 consists of straight line segments

Section 1.3 Insurance and Utility

, ..


smoother functions, the two properties suggested by observation are u'(w) > 0 and u"(w) < o. The second inequality indicates that u(w) is a strictly concave downward


In discussing insurance decisions using strictly concave downward utility function" mP will make use of one form of Jensen's inequalities. These inequalities state

that for a random variable X and function u(w),

if u"(w) < U, then l:',lUV~·)J s U~ClA]),

(1 '1 ')\

if u"(w) > 0, then E[u(X)] > u(E[X])


Jensen's inequalities require the existence of the two expected values. Proofs of the inequalities are required by Exercise 1.3. A second proof of (1.3.2) is almost immediate from consideration of Figure 1.3.1 as follows.

Proof of Jensen's Inequalities for the Case u'(w) > 0

and u"( w} < 0 y

+ u'(J1) (w - J1)



If E[X] = f-L exists, one considers the line tangent to u(w), Y = U(f-L) + u'(f-L)(W - p.),

at the point (u, u(f-L)). Because of the strictly concave characteristic of u(w), the graph of u(w) will be below the tangent line; that is,


for all values of w. If w is replaced by the random variable X, and the expectation is taken on each side of the mequality (l.3.4), we liave E[u(X)] <' u(fl-).

This basic inequality has several applications in actuanal mathematics. Let us apply Jensen's inequality (1.3.2) to the decision maker's insurance problem as formulated in (1.3.1). We will assume that the decision maker's preferences are such that u'(w) > 0 and u"(w) < O. Applying Jensen's inequality to (1.3.1), we have

u(w - G) = E[ll(W - X)] :::; fI(W - u.). (1.3.5)

Chapter 1 The Economics of Insurance

Because u'(w) > 0, u(w) is an increasing function. Therefore, (1.3.5) implies that w _ G ~ w - u, or G 2: fL with G > fL unless X is a constant. In economic terms, we have found that if u'(w) > 0 and u"(w) < 0, the decision maker will pay an amount greater than the expected loss for insurance. If G is at least equal to the premIUm set by the insurer, there is an opportunity for a mutually advantageous

insurance policy.

Formally we say a decision maker with utility function u(w) is risk averse if, and

only if, u"(w) < O.

We now employ a general utility function for the insurer. We let u/(w) denote the utility of wealth function of the insurer and w/ denote the current wealth of the insurer measured in monetary terms. Then the minimum acceptable premium H for assuming random loss X, from the viewpoint of the insurer, may be determined

The left-hand side of (1.3.6) is the utility attached to the insurer's current position.


The right-hand side is the expected utility associated with collecting premium H and paying random loss X In other words, the insurer is indifferent between the

current position and providing insurance for X at premium H. If the Insurer's utility function is such that u;(w) > 0, u7(w) < 0, we can use jensen's inequality (1.3.2)

along with (1.3.6) to obtain

u/(w/) = E[u/(w/ + H - X)1 ~ ulw/ + H - p.).

Following the same line of reasoning displayed in connection with (1.3.5), we can conclude that H :> p;. If 6, as determined by the decision maker by solving (1.3.5),

is such that G :> H :> fL' an insurance contract is possible. That is, the expected

utility of neither party to the contract is decreased.

A utility function is based on the decision maker's preferences for various distributions of outcomes. An insurer need not be an individual. It may be a partnership, corporation, or government agency. In this situation the determination of u/(w), the insurer's utility function, may be a rather complicated matter. For example, if the insurer is a corporation, one of management's responsibilities is the formulation of a coherent set of preferences for various risky insurance ventures. Ihese preferences may involve compromises between conflicting attitudes toward

risk among the groupS of stockholders.

Several elementary functions are used to illustrate properties of utility functions.

Here we examine exponential, fractional power, and quadratic functions. Exercises 1.6, 1.8, 1.9, 1.10(b), and 1.13 cover the logarithmic utility function.

An exponential utility function is of the form

for all 'W and for a fixed a > 0

and has several attractive features. First,

Section 1.3 Insurance and Utility


u'(w) = ae-aU' > O.


u"(w) = _a2e-aw < O.

Therefore, u(w) may serve as the utility function of a risk-averse individual. Third,

is essentially the same as finding the moment generating function (m.g.f.) of X. In

t is expreSSIOn,

Mx(t) = E[etX]

denotes the m.g.f. of X. Fourth, insurance premiums do not depend on the wealth of the decision maker. This statement is verified for the insured by substituting the exponential utility function into (1.3.1). That is,

log MAO')


and G does not depend on w.

The verification for the insurer is done by substituting the exponential utility function with parameter 0'1 into (1.3.6):

_C-a[lI'l == E[ -e -a[(IC/+H-.,\O],

Example 1.3.1

A decision maker's utility function is given by u(w) = _e-511'. The decision maker has two random economic prospects (gains) available. The outcome of the first, denoted by X, has a normal distribution with mean 5 and variance 2. Henceforth, a statement about a normal distribution with mean j.1 and variance (J2 will be abbreviated as N(p, a2) The second prospect denoted by Y, is distributed as

N(6, 2.5). Which prospect will be preferred?

We have

E[ll(X)] = E[ _e-5X]

-Mx(-5) = _el-5(5)+(5")(2)(2]


Chapter 1 The Economics of Insurance



E[u(Y)l = E[ -e-5Yl

= -My(-5) = _el-5(6)+(52)(25)/2]



1 > E[u(Y)]

and the distribution of X is preferred to the distrib1ltion of Y.

In Example 1.3.1 prospect X is preferred to Y despite the fact that j.1x = 5 < j.1y = 6.

Since the decision maker is risk averse, the fact that the distribution of Y is more diffuse than the distribution of X is weighted heavily against the distribution of Y in assessing its desirability. If Y had a N(6, 2.4) distribution, E[u(Y)] = -1 and the decision maker would be indifferent between the distributions of X and Y.

The family of fractional power utility functions is given by

u(w) - w"Y

w > 0, 0 < 'Y < 1.

A member of this family might represent the preferences of a risk-averse decision

maker since

u'(w) = 'YW-y-l > 0


u"(w) = 'Y('Y - 1)wr2 < O.

In this family, premiums depend on the wealth of the decision maker in a manner that may be sufficiently realistic in many situations.

has wealth of w = 10 and faces a random loss X with a uniform distribution on (0, 10). What is the maximum amount this decision maker will pay for complete

insurance against the random loss?


Substitnting into (1.3.1) we have

\,1}0 - G = E[\I10 - X]

x 10 I dx


= -2(10 - X?/2110

3(10) 0

= ~ To 3 V ,


Section 13 Insurance and Utility

The decision maker is risk averse and has u'(w) > O. Following the discussion of (1.3.5), we would expect G > E[X], and in this example G = 5.5556 > E[X] = 5.~

The family of quadratic utility functions is given by

u(w) = w - aw2 w < (2a)-I, a > O.

A member of this family might represent the preferences of a rIsk-averse decision maker since u"(w) - -2a. While a quadratic utility function is convenient because

decisions depend only on the first two moments of the distributions of outcomes under consideration, there are certain consequences of its use that strike some peo-

ple as being unreasonable. Example 1.3.3 illustrates one of these consequences.

A decision maker's utility of wealth function is given by

U(w) = w - 0.01w2

1:V < 50.

The decision maker will retain wealth of amount w with probability p and suffer a financial loss of amount c with probability 1 p. For the values of 'W, c, and p

exhibited in the table below, find the maximum insurance premium that the deci-

sion maker will pay for complete insurance. Assume c :::; w < 50.


For the facts stated, (1.3.1) becomes

utu: - G) = pu(w) + (1 - p)u(w - c), (ll' - G) - O.Ol(w - G)2 = p(w - 0.01(2)

+ (1 - H(w - c) - O.Ol(w - cf]·

For given values of w, p, and c this expression becomes a quadratic equation. Two

solutions are shown.

Wealth Loss Probability Insurance Premium
w c P G
10 10 0.5 5.28
20 10 05 5.37 In Example 1 3.3, as anticipated, G is greater than the expected loss of 5. However,

the maximum insurance premium for exactly the same loss distribution increases with the wealth of the decision maker. This result seems unreasonable to some 'Nho

anticipate that more typical behavior would be a decrease in the amount a decision

maker would pay for insurance when an increase in wealth would permit the decision maker to absorb more of a random loss. Unfortunately, a maximum insurance premium that increases with wealth is a property of quadratic utility functions. Consequently, these utility functions should not be selected by a decision maker who perceives that his ability to absorb random losses goes up with increases in wealth.

Chapter I I he Economics of InsuranCe


....... --------~~~~--------

If we rework Example 1.3.3 using an exponential utility function, we know that the premium G will not depend on w, the amount of wealth. In fact, if u(w) = _e-OOh(', it can be shown that G = 5.12 for both w = 10 and w = 20.

The probability that a property will not be damaged in the next period is 0.75.

The probability density function (p.d.f.) of a positive loss is given by

f<x) O.2S(O.Ole 001X) x > 0

The owner of the property has a utility function given by

Calculate the expected loss and the maximum insurance premium the property owner will pay for complete insurance.


The expected loss is gIven by

ElX] 0.75(0) + 0.25 Jo x(O.Ole o.oL\)dx

= 25.

We apply (1.3.1) to determine the maximum premium that the owner will pay for complete insurance. This premium will be consistent with the property owner's

preferences as summarized in the utility function:

0.75u(w) + C" u(w

11(71' G)



e ll.l1l15(,(' G)

0.751' 0.005H>

0.25 J 0 e 0.C05(,( '\O.Ole 001' )dJ,

eonu'l. _ 0.75 + (0.25)(2) = 1.25,

G = 200 log 1.25 = 44.63.

Therefore, in accord with the property owner's preferences, he will pay up to

44.63 25 - 19.63 in excess of the expected loss to purchase insurance covenng all losses In the next per iod. ~

In Example 1.3.5 the notion of insurance that covers something less than the complete loss is introduced. A modification is made in (1.3.1) to accommodate the fact that losses are shared by the decision maker and the insurance system.


The property owner in Example 1.3.4 is offered an insurance policy that will pay

1 /2 of any loss during the next period. The expected value of the partial loss

Section 13 Insurance and Utility

payment is E[X /2] = 12.50. Calculate the maximum premium that the property owner will pay for this insurance.


Consistent with his attitude toward risk, as summarized in his utility function, the premi1lm is determined from

O.75u(w G) + fo" II (W G ;) f(x)dx

= O.7Su(w) + Jex u(w - x)f(x)dx.


The left-hand side of this equation represents the expected utility with the partial insurance coverage. The right-hand side represents the expected utility with no insurance. For the exponential utility function and p.d.f. of losses specified in Example 1.3.4, it Can be shown that G = 28.62. The property owner is willing to pay up to G jJo - 28.62 12.50 16 12 more than the expected partial loss for the

partial insurance coverage. T

1.4 Elements of Insurance

Individuals and organizations face the threat of financial loss due to random events. In Section 1.3 we saw how insurance can increase the expected utility of a decision maker facing such random losses. Insurance systems are unique in that the alleviation of financial losses in which the number, size, or time of occurrence is random is the primary reason for their existence. In this section we review some of the factors influencing the organization and management of all. insurance system.

All. insurance system can be organized only after the identification of a class of situatIons where random losses lIlay OCCut. The word ralLdom is taken to mean,

along with other attributes, that the frequency, size, or time of loss is not under the

control of the prospective insured. If such control exists, or if a claim payment exceeds the actual financial loss, all. incentive to incur a loss will exist. In such a situation, the assumptions under which the insurance system was organized will become invalid. The actual conditions under which premiums are collected and claims paid will be different from those assumed in organizing the system. The system will not achieve its intended objective of not decreasing the expected utilities of both the inSllfPd and the insurer

Once a class of insurable situations is identified, information on the expected

utilities and the loss-generating process can be obtained. Market research in insurance can be VIewed as all. effort to learn about the utility functions, that is, the risk

preferences of consumers.

The processes generating size and time of loss may be sufficiently stable over time so that past information can be used to plan the system. When a new insurance system is organized, directly relevant statistics are not often available. However, enough ancillary information from similar risk situations may be obtained to

Chapter 1 The Economics ot Insurance


identify the risks altd to plOvide prelirrrrrrarv estimates of the probability distri-

butions needed to determine premiums. Because most insurance systems operate under dynamic conditions, it is important that a plan exist for collecting and analyzing insurance operating data so that the insurance system can adapt. Adaptation in this case may mean changing premiums, paying an experience-based dividend or premium refund, or modifying future policies.

In a competitive economy, market forces encourage insurers to price short-term policies so that deviations of experience from expected value behave as indepelL-

dent random variables. Deviations should exhibit no pattern that might be exploited by the insured or insure! to pi oduce consistent gams. Such consistent

deviations would indicate inefficiencies in the insurance market.

As a result, the classification of risks into homogeneous groups is an important function within a market-based insurance system. Experience deviations that are random indicate efficiencv or equity in classification. In a competitive insurance market, the continual interaction of numerous buyers and sellers forces experimen-

tatioo with classificatIon systems as the market participants attempt to take advantage of perceived patterns of deviations. Because insllTance losses may be relatively

rare events, it is often difficult to identify nonrandom patterns. The cost of classification information for a refined classificatioll system also places a bound on ex-

perimentation in this area.

For insurance systems organized to serve groups rather than individuals, the issue is no longer whether deviations in insurance experience are random for each individual. Instead, the question is whether deviations in group experience are random. Consistent deviations in experience from that expected would indicate the need for a revision in the system.

Group insurance decisions do not rest on individual expected utility comparisons Instead, group insurance plans are based 00 a collective declSlon on whether

the system increases the total welfare of the group. Group health insurance Pro-

vidirrg benefits for the employees of a firm is an example.

1.5 Optimal Insurance

The ideas outlined in Sections 1.2, 1.3, and 1.4 have been used as the foundation of an elaborate theory for guiding insurance decision makers to actions consistent

with their preferences. In this section we present one of the main results from this theory and review many of the ideas introduced so far

A decision maker has wealth of amount w alld faces a loss in the next perIod.

This loss is a random variable X. The decision maker can buy an insurance contract that will pay I(x) of the loss x. To avoid an incentive to incur the loss, we assume that all feasible insurance contracts are such that 0 -s I(x) :s x. We make the

Section 7.5 Optima/Insurance

simplIfymg assumption that all feasible insurance contracts with E[1(K)] 13 can be

purchased for the same amount P.

The decision maker has formulated a utility function II(W) that is consistent with his preferences for distributions of outcomes. We assume that the decision maker is risk averse, u"(w) < O. We further assume that the decision maker has decided on the amount, denoted by P, to be paid for insurance. The question is: which of

. the insurance contracts from the class of feasible contracts with expected claims, 13, and premium, P, shonld be pnrchased to maximize the expected utility of the de-

cision maker?

One subclass of the class of feasible insurance contracts is defined as follows:

I (x) = {o

d X - d

x<d x :2: d.

(1.5.1 )

This class of contracts is characterized by the fact that claim payments do not start until the loss exceeds the deductible amount d. For losses above the deductible

amount, the excess is paid under the terms of the contract. Ihls type of contract is sometimes called stop-loss or excess-ot-Ioss insurance, the choice depending on

the application.

In the problem discussed in this section the expected claims are denoted by 13·

In (1.5.2) the symbol ]tx) denotes the p.d.f. and the symbol F(x) denotes the distribution function (d.f.) associated with the random loss X:

13 = f' (x - d)f(x)dx



13 = t [1 - F(x»)dx.


Equation (1.5.2B) is obtained from (1.5.2A) by integration by parts. When 13 is given, then (1.5.2) prOVIdes explicit equations for the correspolldillg deductible, delLoted

by d". In Exercise 1.17, it is shown that d* exists and is unique.

The main result of this section can be stated as a theorem.

Chapter 7 The Economics of Insurance


Theorem 1.5.1 is an important result and illustrates many of the ideas developed in this chapter. However, it is instructive to consider certain limitations on its applicability. First, the ratio of premium to expected claims is the same for all available contracts. In fact, the distributions of the random variables J(X) can be very differ-

ent, and the provIsIOn for nsk in the premiuIlI usually depel1ds on the characteristics of the distribution of I(X). Second, in Theorem 1.5.1, it is assumed that the

premium P is fixed by a budget constraint. Alternatives to amount P are not considered. In Exercise 1.22, relaxation of the budget constraint is considered. Third,

while the theorem indicates the form of insurance, it does not help to determine

the amount P to spend. In the theorem, P is fixed.

1.6 Notes and References

Definitions and principles of actuarial science can be found in "Principles of Actuarial Science" (SOA Committee on Actuarial Principles 1992).

The role of risk in business was developed in a pioneering thesis by Willett (1951).

Borch (1974) has published a series of papers applying utility theory to insurance questIOns. DeGroot (1970) gi\ es a contplete development of utility theory starting

from basic axioms for consistency among preferences for various distributions of outcomes. DeGroot and Borch both discuss the historically important St. Petersburg paradox, outlined in Exercise 1.1. A paper by Friedman and Savage (1948) provides many insights into utility theory and human behavior.

Pratt (1964) has studied (1.3.1) and derived several theorems about premiums

and utility functions. Exercise 1.10, whIch uses two rough approxrrrratiorrs: is related to one of Pratt's results.

Theorem 1.5.1 on optimal insurance was proved by Arrow (1963) in the context

of health insurance. The theorem in Exercise 1.21, in which the goal of insurance is to minimize the variance of retained losses, was the subject of papers by Borch (1960) and Kahn (1961). The use of the variance of losses as a measure of stability is discussed by Beard, Pentikainen, and Pesonen (1984). Exercise 1.23 is based on their discussion.


Section 1. 6 Noles and Refel enccs



If u"(w) < 0 for all w in [a, b], then for wand z in [a, b],

II(W) - II(Z) ::::; (w - z)u'(z).



The lemma may be established with the aid of Figme ] 3] I Ising the point

slope form, a line tangent to u(w) at the point (z, u(z)) has the equation y - u(z) = fl'(Z)(W - z) and is above the graph of the function u(w) except at the point (z, u(z)). Therefore,

!leW) - u(z) ::::; u'(z)(w - z).

Figure 1.3.1 shows the case u'(w) > o. The same argument holds for u'(w) < o.

In Exercise 1.20 an alternative proof is required.

Proof of Theorem 1.5.1:

Let J(x) be associated with an insurance policy satis~ying the hypothesis of tlte

theorem. Then from the lemma,

uiu: - x + lex) - P) - u(w - x + ld'(x) - P) ::::; [lex) - IAx)]u'(w - x + Id'(x) - P).

In addition, we claim


[lex) IAx)]u'(w x + 1d'(x) P)

::::; [l(x) ld'(x)]u'(w d* P).

To establish inequality (l.A.3), we must consider three cases:


Case 1. If.(x) = lex)

In this case equality holds, (1.A.3) is 0 on both sides.

Case II. lAx) > lex)

In this case ld'(x) > 0 and from (1.5.1), x - ld'(x) = d", Therefore, equality holds with each side of (1.A.3) equal to [lex) - IAx)]u'(w - £1* - P).

II. d'(X) < (x)

In this case lex) - ld.(.t") > O. From (1.5.1) we obtain lAx) - x 2: -d*

and l<i'(x) - X - P 2: +d" - P. Therefore,

u'(to - x + Id'(X) - P) ::::; II'(LO - d* - P)

since the second derivative of u(x) is negative and u'(x) is a decreasing function.

Chapter 1 The Economics of Insurance


Therefore, in each case

[l(x) - ld'(X)]U'(W - X + lAx) - P) ::; [l(x) - l,t*(x)]U'(W - P - d*), establishing inequality (1.A.3).

Now, combining inequalities (1.A.2) and (1.A.3) and taking expectations, we have

E[u(w - X + leX) - P)] - E[u(w - X + ld'(X) - P)]

::; E[l(X) - ld'(X)]U'(w - d* - P) = (13 - (3)u'(w - d* - P) = o.


E[u(w - X + leX) - P)] ::; E[u(w - X + Id'(X) - P)]

and the expected utility will be maximized by selecting Id.(x), the stop-loss policy . •


Section 1.2

1.1. Assume that a decision maker's current wealth is 10,000. Assign u(O) = -1

and u(10,OOO) - o.

a. When facing a loss of X with probability 0.5 and remaining at current wealth with probability 0.5, the decision maker would be willing to pay up to G for complete insurance. The values for X and G in three situations are given below.

10 000 6 000

3300 1700

Determine three values on the decision maker's utility of wealth function u.

b. Calculate the slopes of the four line segments joining the five points determined on the graph u(w). Determine the rates of change of the slopes from segment to segment.

c. Put yourself in the role of a decision maker with wealth 10,000. In addition

to the given values of 1I(0) and u(10,OOO), elicit three additional values on your utility of 'Nealth function u;

d. On the basis of the five values of your utility function, calculate the slopes

and the rates of change of the slopes as done in part (0).

1.2. St. Petersburg paradox: Consider a game of chance that consists of tossing a coin until a head appears. The probability of a head is 0.5 and the repeated trials are independent. Let the random variable N be the number of the trial on which the first head occurs.


a. Show that the probability function (p.f.) of N is given by


!(n) = 2

n = 1,2,3, ....

b. Find E[N] and Var(N).

c. If a reward of X = 2''1 is paid, prove that the expectation of the reward does not exist.

d. If this reward has utility u(w) = log 10, find E[u(X)].

Section 1.3

1.3. Jensen's inequalities:

a. Assume u"(w) < 0, E[X] = J.1, and E[u(X)] exist; prove that E[u(X)] ::S u(J.1). [Hint: Express u(w) as a series around the point w = J.1 and terminate the expansion with an error term involving the second derivative. Note that Jensen's inequalities do not require that u'(w) > 0.]

b. If lI"(w) > 0, prove that E[u(X)] 2:: u(J.1).

c. Discuss Jensen's inequalities for the special case u(w) = w2. What is

E[lI(X)] - u(E[X])?

1 4 If a utility function is sllch that u'(w) > 0 and u"(w) > O. use (1.3.1) to show

G ::S J.1. A decision maker with preferences consistent with u"(w) > 0 is a risk

1.5. Construct a geometric argument, based on a graph like that displayed in Figure 1.3.1, that if ll'(W) < 0 and u"(w) < 0, then (1.3.4) follows.

1.6. Confirm that the utility function u(w) = log 10, 10 > 0, is the utility function of a decision maker who is risk averse for w > O.

1 7 A utility function is given by

1(1 < 100

W 2:: 100.

a. Is u'(w) 2: O?

b. For what range of ((' is ll"(W) < O?

1.8. If one assumes, as did D. Bernoulli in his comments on the St. Petersburg paradox, that utility of wealth satisfies the differential equation

10 > 0, k > 0,

confirm that ll(W) = k log w + c.

1.9. A decision maker has utility function ll(W) = k log w. The decision maker has wealth w, 10 > 1, and faces a random loss X, which has a uniform distribution on the interval (0, 1). Use (1.3.1) to show that the maximum insurance premium that the decision maker will pay for complete insurance is

Chapter 1 The Economics of Insurance



G = w - i '

e(w - 1)'"

1.10. a. In (1.3.1) use the approximations

uito - G) == u(w - u.) + (p, - G)u'(w - u ),

11(10 - x) == u(w - u.) + (u - x)u'(w - u.) + :2 (f.L - X)2U"(W - u)

and derive the following approximation for G:

G - 1 u"(w - f.L) 2

- f.L - - a

- 2 u'(w - u.) .

b. If u(w) = k log 10, use the approximation developed in part (a) to obtain 1 a2


2 (w - u.)

1.11. The decisioIl maker has a utility function u(w) e "l<' and is faced with a

random loss that has a chi-square distribution with n degrees of freedom. If

0< a < 1/2, use (1.3.1) to obtain an expreSSIOn for G, the maximum insurance

premillm the decision maker will pay, and prove that G > n f.L.

1.12. Rework Example 1.3.4 for

a. u(w) = _e-w/400

b. u(w) = _e-w/150•

1.13. -a. An insurer with net worth 100 has accepted (and collected the premium for) a risk X w itl! tl re folio V\i ing probability distribution:

Pr(X 0) Pr(X 51)



What is the maximum amount G it should pay another insurer to accept 100% of this Joss? Assume the first insurer's utility function of wealth is

u(w) = log w.

b. An insurer, with wealth 650 and the same utility function, u(w) = log w,

is considering accepting the above risk. What is the minimum amount H this insurer would accept as a premium to cover 100% of the loss?

1.14. If the complete insurance of Example 1.3.4 can be purchased for 40 and the

50% COinsurance of Example 1.3.5 can be purchased fOI 25, the purchase of which insurance maximizes the property owner's expected utility?

Section 1.4

1.15. A hospital expense policy is issued to a group consisting of n individuals.

The policy pays B dollars each time a member of the group enters a hospital.


The group is not homogeneous with respect to the expected number of hospital admissions each year. The group may be divided into r subgroups. There are n, individuals in subgroup i and l; n, = n. For subgroup i the number of annual hospital admissions for each member has a Poisson distribution with parameter A" i = I, 2, ... , r. The number of annual hospital admissions for members of the group are mutually independent.

a. SLov~ that the expected claims payment in one year is



2: niAi ~ = _1 __


b. Show that the number of hospital admissions in 1 year for the group has a Poisson distribution with parameter n ~.

1.16. Perform the integration by parts indicated in (1.5.2). Use the fact that if E[X]

exists, if and only if, lim x[I - F(x)] = O.


1.17. a. Differentiate the right-hand side of (1.5.2B) with respect to d.

b. Let 13 be a number such that a < 13 < E[X]. Show that (1.5.2) has a unique solution d".

1.18. Let the loss random variable X have a p.d.f. given by

f(X) O.Ie

a Calculate E[X] and Var(X).

b. If P = 5 is to be spent for insurance to be purchased by the payment of the pure prerrmrm. 5ho'W that


[(x) = 2:



x < d

x > d, where d - 10 log 2.

both represent feasible insurance poliCies with pure premium P = 5. [(x)

is called proportional insurance.

1.19. The loss random variable X has a p.d.f. given by 1 f(x) = 100

a < x < 100.

Chapter 1 The Economics of Insurance


a. Calculate E[X] and Var(X).

b. Consider a proportional policy where

I(x) = kx and a stop-loss policy where

lAx) { 0

o < k < 1,



x;:::: d.

Determine k and d such that the pure premium in each case IS P - 12.5.

c. Show that Var[X T(X)] > Var[X Id(x)].


1.20. Establish the lemma by using an analytic rather than a geometric argument. [Hint: Expand u(w) in a series as far as a second derivative remainder around the point z and subtract u(z).]

1.21. Adopt the hypotheses of Theorem 1.5.1 WIth respect to 13 and insurance COIl-

tracts T(x) and assume E[X] 1.1. Prove that

Var[X - I(X)] = E[(X - leX) - I.L + [3)2]

is a minimum when l(x) = ld'(x). You will be proving that for a fixed pure

premmm, a stop-loss insurance contract will minimize the variance of Ie tained claims. [Hint: we may follow the proof of Theorem 1.5.1 by first proving that x2 - Z2 ;:::: (x - z)(2z) and then establishing that

[x - l(x)f - [x - IAx)j2 ;:::: [I,t*(x) - l(x)][2x - 2Id'(x)]

;:::: 2[Id'(x) - l(x)]d*.

The final inequality may be established by breaking the proof into three cases.

Alternatively, by proper choice of wealth level and utility function, the result

of this exercise is a speCIal case of rheorem 1.5.1.]

1.22. Adopt the hypotheses of Theorem 1.5.1, except remove the budget constraint; that is, assume that the decision maker will pay premium P, 0 < P ~ E[X] = I.L, that will maximize expected utility. In addition, assume that any feasible insurance can be purchased for its expected value. Prove that the optimal insurance is lo(x). This result can be summarized by stating that full coverage is optimal in the absence of a budget constraint if insurance can be purchased

for its pure premium. [Hint· lIse the lemma with the role of w played bv w

x + I(x) - P and that of z played by IV - X + Io(x) - E[X] = w - I.L. Take

expectations and establish that E[u(zv X) I I(X) P] -< !I(w foL)·]

1.23. Optimality properties of stop-loss insurance were establIshed In rheorem 1.5.1 and Exercise 1.21. These results depended on the decision criteria, the constraints, and the insurance alternatives available. In each of these developments, there was a budget constraint. Consider the situation where there is a

risk constraint and the price of insurance depends on the insurance risk as measured by the variance.

(i) The insurance premium is E[I(X)] + f(Var[I(X)]), where few) is an increasing function. The amount of f(Var[I(X)]) can be interpreted as a security loading.

(ii) The decision maker elects to retain loss X - leX) such that Var[X - I(X)] F > O. This requirement imposes a risk rather than a budget constraint.

The constant is determined by the degree of risk aversion of the decision

maker. FIxmg the accepted variance, and then optimizing expected results, is a decision criterion in investment portfolio theory.

(iii) The decision maker selects l(x) to minimize f(Var[I(X)]). The objective is to minimize the security loading, the premium paid less the expected insurance payments. Confirm the following steps:

a. Var[I(X)] = V + Var(X) - 2 Cov[X, X - I(X)].

b. The l(x) that minimizes Var[I(X)] and thereby f(Var[I(X)]) is such that the correlation coefficient between X and X - l(x) is 1.

c. It is knovm that if hvo random variables Wand Z have correlation

coefficient 1, then Pr{W = aZ + b, where a > O} = 1. In words, the

probabIlIty of thelI [oint dIstnbuhon IS concentrated on a lme of positive slope. In part (b), the correlation coefficient of X and X - l(X)

was found to be 1. Thus, X - leX) = aX + b, which implies that I(X) - (1 a)X b. To be a feasible insurance, 0 <: I(x) <: x or 0 <:

(1 - a)x - b :s:; x. These inequalities imply that b = 0 and 0 :s:; 1 - a :s:; 1 and 0 :s:; a :s:; 1.

d. To determine a, set the correlation coefficient of X and X - I(X) equal to 1, or equivalently, their covariance equal to the product of their standard deviations. Thus, show that a = vv / Var(X) and thus that the insurance that minimizes f(Var[X]) is leX) = [1 - Vv / Var(X)]X.

Chapter 1 The Economics of Insurance




2.1 Introduction

In Chapter 1 we examined how a decision maker can use insurance to reduce the adverse financial impact of some types of random events. That examination

was quite general. The decision maker could have been an individual seeking pro-

tection against the loss of property, savings, or income. The deClsion maker could have been an organization seeking protection against those same types of losses. In fact, the organization could have been an insurance company seeking protection against the loss of funds due to excess claims either by an individual or by its portfolio of insureds. Such protection is called reinsurance and is introduced in this chapter.

The theory in Chapter 1 requires a probabilistic model for the potential losses.

Here we examine one of two models commonly used in insurance pricing, resenT-

ing, and reinsurance applications.

For an insuring organization, let the random loss of a segment of its risks be denoted by s. Then 5 is the random variable for which we seek a probability distribution. Historically, there have been two sets of postulates for distributions of s. The individual risk model defines

5 = x, + X2 + ... + XII


where XI IS the toss on Insured UnIt I and n IS the number of nsk units Insured. Usually the X's are postulated to be independent random variables, because the

mathematics is easier and no historical data on the dependence relationship are needed The other model is the collective risk model described in Chapter 12.

The individual risk model in this chapter does not recognize the time value of money. This is for simplicity and is why the title refers to short terms. Chapters 4- 11 cover models for long terms.

Chapter 2 Individual Risk Models for a Short Term



· In this chapter we discuss only closed models; that is, the number of insured units n in (2.1.1) is known and fixed at the beginning of the period. If we postulate about migration in and out of the insurance system, we have an open model.

2.2 Models for Individual Claim Random Variables

ce ts with a life insurance product. In a one-year term

life insurance the insurer agrees to pay an amount b if the insured dies wit in a year of pohcy issue and to pay nothing if the insured survives the year. The prob-

ability of a claim during the year is denoted by q. The claim random variable, X, has a distribution that can be described by either its probability function, p.f., or its distribution function, d.f. The p.f. is

1 - q fx(x) = Pr(X = x) = q


x = 0

x = b elsewhere,


and the d. f. is


Fx(x) Pr(X <: x) .,

x<O O<x<h


x> b.

From the p.f. and the definition of moments,

E[X] = bq, E[X2] = b2q,



Var(X) b2q(1 q).

(2.2.4) .

e formulas can also be obtained by writing

x = Ib


where b is the constant amount payable in the event of death and I is t e ran am variable that is 1 for the event of death and ° otherwise. Thus, Pr(I = 0) = 1 - q and Pr(I = 1) = q, the mean and variance of I are q and q(l - q), respectively, and the mean and variance of X are bq and b2q(1 - q) as above.

The random variable I with its 10, I} range is widely applicable in actuarial models. In probability textbooks it is called an indicator. Bernoulli random variable, or

binomial random variable for a single trial. We refer to it as an indicator for the

sake of brevity and because It mdicates the occtrrrenee. I I, or nonocourence, I O. of a given event.

We now seek more general models in which the amount of claim is also a random variable and several claims can occur in a period. Health, automobile, and other property and liability coverages provide immediate examples. Extending (2.2.5), we postulate that

X = IB



Section 2.2 iVlOo'els (0/ individoa! Claim Random Vsriables

where X is the claim random variable for the period, B gives the total claim amount incurred during the period, and I is the indicator for the event that at least one claim has occurred. As the indicator for this event, 1 reports the occurrence (l = 1) or nonoccurrence (I = 0) of claims in this period and not the number of claims in the period. Pr(l = 1) is still denoted by q.

Let us look at several situations and determine the distributions of I and B for a

model. First, consider a I-year term life insurance paying an extra benefit in case

of accidental death. To be specific, if death is accidental, the beneht amount IS 50,000. For other causes of death, the benefit amount is 25,000. Assume that for the

age, health, and occupation of a specific individual, the probability of an accidental death within the year is 0.0005, while the probability of a nonaccidental death is 0.0020. More succinctly,

Pr(l = 1 and B = 50,000) = 0.0005


PrU I and B 25,000) 0.0020.

Summillg over the possible values of B, we ha ve

and then

Pr(I = 0) = 1 - Pr(l = 1) = 0.9975.

The conditional distribution of B, given 1 = I, is

P B = 000[1 = = Pr(B = 25,000 and I = 1) = 0.0020 = 08

r( 25, 1) Pr(I = 1) 0.0025' ,

Pr(B = 50,000 and I = 1) 0.0005

Pr(B 50,000[1 I) 02

Pr(I = 1) 0.0025' .

Let us now consider an automobile insurance providing collision coverage (this indemnifie~ the ovmer for collision damage to his car) above a 250 deductible up

to a maximum claim of 2,000. For illustrative purposes, assume that for a particular individual the probability of one claim in a period is 0.15 and the chance of more than one claim is 0:

Pr(l = 0) = 0.85, Pr(I = 1) = 0.15.

This unrealistic assumption of no more than one claim per period is made to simplify the distribution of B. We remove that assumption in a later section after we

discuss the distribution of the sum of a number of claims. Since B is the claim incurred by the insurer, rather than the amount of damage to the car, we can infer

two characteristics of I and B. First, the event 1 = 0 includes those collisions in which the damage is less than the 250 deductible. The other inference is that B's distribution has a probability mass at the maximum claim size of 2,000. Assume

Chaoter 2 Individual Risk Models for a Short Term


this probability mass is 0.1. Furthermore, assume that claim amounts between 0 and 2,000 can be modeled bya continuous distribution with a p.d.f. proportional to 1 - x /2,000 for 0 < x < 2,000. (In practice the continuous curve chosen to represent the distribution of claims is the result of a study of claims by size over a recent period.) Summarizing these assumptions about the conditional distribution of B, given I-I, we have a mixed distribution with positive density from 0 to

2,000 and a mass at 2,000. This is illustrated in Figure 2.2.1. The d.f. of this conditional distribution is



Pr(B ::; xiI = 1) =

0.9 [1 - (1 __ x_ 2,000


o < x < 2,000

X 2: 2,000.

Distribution Function for B, given I 1


1.0 0.9

__ -L--------------------------~--------x



We see in Section 24 that the moments of the claim random variable, X, in

particular the mean and variance, are extensively used. For this automobIle insurance, we shall calculate the meall and the variance by two methods. First, we derive

the distribution of X and use it to calculate E[X] and Var(X). Letting Fx(x) be the d.f. of X, we have

Fx(x) = Pr(X ::; x) = Pr(IB ~ x)

= Pr(IB ::; xiI = 0) Pr(I = 0)

+ Pr(IB < xiI 1) Pr(I 1).


Fx(x) - 0(0.85) + 0(0.15) - O.

For 0 ::; x < 2,000,

Fx(x) = 1(0.85) + 0.9 [1 - (1 - 2,~00YJ (0.15).


Sectioll 2.2 Models for Individual Claim Random Variables

For x ::::: 2,000,

FAx) = 1(0.85) + 1(0.15) = 1.

This is a mixed distribution. It has both probability masses and a continuous part as can be seen in its graph in Figure 2.2.2.

Distribution Function of X - IB






Corresponding to this d.f. is a combination p.f. and p.d.f. given by

Pr(X = 0) = 0.85, Pr(X = 2,000) = 0.015


with p.d.f.


F ~(x) = 0.000135 (1 - 2,;00)

o < x < 2,000



Moments of X can then be calculated by


E[Xk] = 0 x Pr(X = 0) + (2,000)k X Pr(X = 2,000) + 0 xkf.'{(x) dx,



E[X] = 120


E[X2] = 150,000.


Var(X) = 135,600.

There are some formulas relating the moments of random variables to certain conditional expectations. General versions of these formulas for the me~n and variance are

E[W] = E[E[WI V]]


Chapter 2 Individual Risk Models for a Short Term



Var(W) = Var(E[W\V]) + E[Var(W\V)].


In these equations we think of calculating the terms of the left-hand sides by direct use of W's distribution. In the terms on the right-hand sides, E[W\V] and Var(W\V) are calculated by use of ¥V's conditional distribution for a given value of V. These

components are then functions of the random variable V, and we can calculate their moments by use of V's distribution

In many actuarial models conditional distributlOns are used. This makes tile for mulas above directly applicable. In our model, X = IB, we can substitute X for W

and I for V to obtain

E[X] = E[E[X\I]]



Var(X) Var(£[X\I]) + E[Var(X\l)]


IJ. E[B\I I],


(J2 = Var(B\I = I),


and look at the conditional means

E[X\I = 0] = 0



E[X\I = 1] = E[B\I = 1] = IJ..


Formulas (2.2.16) and (7 7 17) define E[XII] as a function of I, which can be written

bv the formula



E[E[X\I]] = IJ.E[I] = IJ.q



Var(E[X\I]) = 1J.2 Var(T) = 1J.2q(1 - q).


Since X 0 for I 0, we have

Var(XII 0) a


For I I, '.'Ie have X Band

Var(X\I = 1) = Var(B\I = 1) = (J2.

Formulas (2.2.21) and (2.2.22) can be combined as Var(X\I) = (J2I.




Section 2.2 Models for Individual Claim Random variables



Substituting (2.2.19), (2.2.20), and (2.2.24) into (2.2.12) and (2.2.13), we have

E[X] = fJ-q (2.2.25)


Let us now apply these formulas to calculate E[X] and Var(X) for the automobile

insurance in Figure 2.2.2. Since the p.d.f. for B, given I = 1, is

{ (x)

0.0009 1 ---

fBI/(xI1) = 0 2,000

o < x < 2,000


with Pr(B - 2,0001 I - 1) ° 1, we have

fJ- Jo 0.0009 xl

2,;00) dx + (0.1)(2,000) 800,

_,0 °

E[B21 I = 1] = 0.0009 x2

dx + (0.1 )(2,000l = 1,000,000,


cr2 = 1,000,000 - (800f = 360,000.

Finally, with q = 0.15 we obtain the following from (2.2.25) and (2.2.26):

E[X] = 800(0.15) = 120


Var(X) (800f(O.15)(O.85) + (360,000)(0.15)

There are other possible models for B in different insurance situations. As an example, let us consider a model for the number of deaths due to crashes during an airline's year of operation. We can start with a random variable for the number of deaths, X, on a single flight and then add up a set of such random variables over the set of flights for the year. For a single flight, the event I = 1 will be the event of an accident during the flight. The number of deaths in the accident, B,

will be modeled as the product of two random variables, Land Q, where L is the load factor, the Dumber of persons on board at the time of the cra~h, and Q i~ the

fraction of deaths among persons on board. The number of deaths B is modeled in

this way since separate statistical data for the distributions of Land Q may be more readily available than are total data for B. We have X = ILQ. While the fraction of passengers killed in a crash and the fraction of seats occupied are probably related, Land Q might be assumed to be independent as a first approximation.

Chaplel 2 llidividual Risk lvTodels for a Short Term


2.3 Sums of Independent Random Variables

In the individual risk model, claims of an insuring organization are modeled as the sum of the claims of many insured individuals.

The claims for the mdIViduals are assumed to be iItdependent in most applica tions. In this section we review two methods for determining the distribution of

the sum of independent random variables. First, let us consider the sum of two

random variables, 5 X + 1; with the sample space shown in Figure 2.3.1.

Event {X + Y::s s}


The line X + Y = s and the region below the line represent the event

[S ::s X +

Hence the dJ. of 5 is

F,(5) Pr(S <: s) Pr(X + Y <: s)


For two discrete, non-negative random variables, we can use the law of total probability to write (2.3.1) as

Fs(s) = 2 Pr(X + Y::s slY = y) Prey = y)

011 1/<;;;

= 2 Pr(X::S s - ylY = y) PreY = y).


When X and Yare independent, this last sum can be written


The p.f. corresponding to this d.f. can be calculated by fs(s) = L fx(s - y)fy(Y)·

all yss


For continuous, non-negative random variables the formulas corresponding to (2.3.2), (2.3.3), and (2.3.4) are


SectIon 2.3 Sums of independent Random Variables

Fs(s) = fa' Pr(X :::; s - ylY = y)fy(y) dy, Fs(s) = f' Fx(s - y)Jy(y) dy,





When either one, or both, of X and Y have a mixed-type dlstnbutIon (typIcal in individual risk model applications), the formulas are analogous but more complex.

For random variables that may also take on negative values, the sums and integrals in the formulas above are over all y values from -x to +x.

In probability, the operation in (2.3.3) and (2.3.6) is called the convolution of the pair of distribution functions Fx(x) and Fy(y) and is denoted by Fx * Fy. Convolutions can also be defined for a pair of probability functions or probability density functions as in (2 34) and (2 37)

To determiIle the distribution of the sum of more than two random vanables, we can use the convolution process iteratively. For 5 = XI + X2 + ... + XI! where

the X;'s are independent random variables, F, is the d.f. of Xi' and Pk) is the d.f. of XI + X2 + ... + ~, we proceed thus'

F(3) = F3 * F(2) F(~) = F~ * F(3)

Example 2.3.1 illustrates the procedure using probability functions for three discrete

random variables.


The random variables XI' X2, and X3 are independent with distributions defined by columns (1), (2), and (3) of the table below. De;ive the p.f. and d.f. of

So ution:

The notation of the previous paragraph is IIsed in the table'

• Columns (1 )-(3) are given information .

• Column (4) is derived from columns (1) and (2) by use of (2.3.4).

• Column (5) is derived from columns (3) and (4) by use of (2.3.4).

The determination of column (5) completes the determination of the distribution of S. Its d.f. in column (8) is the set of partial sums of column (5) from the top.

Chapter 2 IndividUal Risk Mode/:; f07 a 51i071 Tel/n

. 35

1 \ \


(1) (2)

Nx) f2(X)

(3) Nx)

(8) F (3l(X)

(7) F (2l(X)

o 1 2

0.4. .0.5 O.l~.,\ ''',' ",,0.2 0.2~0.1

0.6. 0.0 0.1

,0.20 -B.23 '0.20 , 0.16\

0.120 0.138 0.140

0.139 0.129 0.115

0.4 0.7 0.9

1.0 1.0 10

0.20 0.43 0.63

0.79 0.90 096

0.120 0.258 0.398

3 4 5

0537 0.666 0.781

0.1 0.1 0.1

·'~b.J.i 0.06

, 0.03·· 0,01

0.869 0.928

-. \

6 7

8 9 10 11 12

0.088 0059

0.036 0.021 0.010 0.004 0.001

1.0 1 0

1.0 1.0 1.0 1.0 1.0

0.99 1.00

1.00 1.00 1.00 1.00 1.00

0.964 0.985 0.995 0.999 1.000

For illustrative purposes we include column (6), the d.f. for column (l ), column (7)

which can be derived directly from columns (2) and (6) by use of (2.3.3), and column (8), derIved Similarly from columns (3) and (7). Column (5) can then be ob-

tained by differencing column (8). ....

We follow with two examples involving continuous random variables.

Let X have a uniform distribution on (0, 2) and let Y be independent of X with a uniform distribution over (0, 3). Determine the d.f. of 5 = X + Y.

Since X and Yare continuous, we use (2.3.6):

o x < 0


f _~1

/,(21) t-


2 1





The X, Y sample space is illustrated in Figure 2.3.2. The rectangular region contains all of the probability for X and Y. The event of interest, X + Y :::: s, has been


Section 2.3 Sums of Indep@ndent Random Variables

Convolution of Two Uniform Distributions y







illustrated in the figure for five values of s. For each value, the line intersects the

y-axIs at s and the line x - 2 at s 2. 'Ihe values of Fs for these fIVe cases are

f' 5 - Y 1 52 ---dy=-

o 2 3 12

J'-2 1 J' s - Y 1 5 - 1

Fs(s) = < 1 - dv + -- - dy = --

LJ 3 - ,-2 2 3' 3

o ::s 5 < 2 line B


5 < 0

line A

2 :::; 5 < 3 line C

Jo • 3 "Y

e 5 - Y 1 A.


12 ~


5 ~ 5

line E.


Consider three independent random variables Xl' X2, X3. For i = 1, 2, 3, Xi has an exponential distribution and E[XJ = 1 / i. Derive the p.d.f. of 5 = X, + X2 + X3 by the convolution process.

!2(X) = 2e-2x f3(X) = 3e-3x

x> 0,

x> O.

Using (2.3.7) twice, we have

Chapter 2 Individual Risk Models for a Short Term




= rx (2e-<x-y) _ 2e-2(X-Y))3e-3y d11

Jo .

= 6e-x fa' e-2y dy - 6e-2x fa' «» dy = (3e-X - 3e-3X) - (6e-2x - 6e-3x)

3e ( 6e 2x + 3e 3<

x> O.

Another method to determine the dIstnbutIon of the sum of random variables is based on tbe uniqueness of the moment generating function (m.g.f.), which, for the

random variable X, is defined by Mx(t) E[etX]. If this expectation IS fImte for all t In an open inter va1 about the origin, then Mx(t) is the only m.g.f. of the distri-

bution of X, and it is not the m.g.f. of any other distribution. This uniqueness can be used as follows. For the sum S = X, + X2 + ... + XII'

Ms(t) = E[etS] = E[et(X1 + X2 + ... + XI1)]


If Xl' X2, ... , XII are independent. then the expectation of the product in (2.3.8) is

equal to

so that

Ms(t) = Mx,(t) Mx,(t) ... Mx,,(t).


Recognition of the unique distribution corresponding to (2.3.9) would complete the determination of S's distribution. If inversion by recognition is not possible, then inversion by numerical methods may be used. (See Section 2.6.)

Consider the random variables of Example 2.3.3. Derive the p.d.L of 5 Xl +

X2 + X3 by recognition of the m.g.f. of S.


By (2.3.9), Ms(t) = (_1_)(_2_)(_33 ), which we write, by the method of 1-t 2-t -t

partial fractions, as


Section 23 Sums of inciepellcient Random Variab';

A 2B 3C

M,(t) = -- + -- + --.

. I-t 2-t 3-t

The solution fur this is A = 3, B = -3, C = 1. But 131 (13 - t) is the moment generating function of an exponential distribution with parameter 13, so the p.d.f. for 5 is


The inverse Gaussian distribution was developed ill the study of stochastic pro-

cesses. Here it is used as the distribution of B, the claim amount. It will have a similar role in risk theory in Chapters 12-14. The p.d.f. and m.g.f. associated with the inverse Gaussian distribution are given by

A(x) = ~ x-3i2 exp[ (I3X2;x a)2] x > 0,

Find the distribution of 5 - Xl + X~ + X3 + ... + X" where the random variables

Xl' Xl' ... r X" are independent and have identical inverse Gaussian distributions.


Using (2.3.9), the m.g.f. of 5 is given by

;vfs(t) = [M\(t)]" = exp [na (1 - J 1 - ~) 1

The m g.f AI(t) can be recognized and shows that 5 has an inverse Gaussian dis-

tribution with parameters Ila and 13. ""

2.4 Approximations for the Distribution of the Sum

The central limit theorem suggests a method to obtain numerical values for the distribution of the sum of independent random variables. The usual statement of the theorem is for a sequence of independent and identically distributed random variables, XI' Xc, ... , with EfXJ = j..L and Var(X,) = cr2• For each II, the distribution of \ Iz (X" - p.) 1 c , where XII = (X, + Xl + ... + XII) 1 n, has mean 0 and vari-

ance 1. The sequence of distributions (11 = I, 2, ... ) is known to approach the standard normal distribution When IJ is large the theorem is applied to approxi-

mate the distribution of X" by a normal distribution with mean f.L and variance cr2! II. Equivalently, the distribution of the sum of the rz raIldom variables is ap-

proximated by a normal distribution with mean 11j..L and variance ncr2. The effectiveness of these approximations depends not only on the number of variables but also on the departure of the distribution of the summands from normality. Many elementary statistics textbooks recommend that 11 be at least 30 for the

Chanter 2 inoividuel Risk Models for a Short Term


approximations to be reasonable. One routine used to generate normally distributed random variables for simulation is based on the average of only 12 independent random variables uniformly distributed over (0, 1).

In many individual risk models the random variables in the sum are not identically distributed. This is illustrated by examples in the next section. The central limit theorem extends to sequences of nonidentically distributed random variables.

To illustrate some applications of the IndIVIdual fIsk model, we use a normal approximation to the distribution of the sum of independent random variables to

obtain numerical answers. If



E[S] = 2: E[Xd,


and, further, under the assumption of independence,


Var(S) = 2: Var(Xk)·

k 1

For an application we need only

• Evaluate the means and variances of the individual loss random variables

• Sum them to obtain the mean and variance for the loss of the insuring organization as a whole

• Apply the normal approximation.

Illustrations of this process follow.

2.5 Applications to Insurance

In this section four examples illustrate the resllits of Section 2.2 and use of the

normal approximation.


A life insurance company issues l-year term life contracts for benefit amounts of 1 and 2 units to individuals with probabilities of death of 0.02 or 0.10. The following table gives the number of individuals 11, in each of the four classes created by a

benefit amount b, and a probability of claim {h.

1 0.02 1 500
2 0.02 2 500
3 0.10 1 300
4 0.10 2 500 40

Section 2.5 Applications to Insurance

The company wants to collect, from thIs population of 1,800 individuals, an amount equal to the 95th percentile of the distribution of total claims. Moreover, it wants each individual's share of this amount to be proportional to that individual's expected claim. The share for individual j with mean E[X) would be (1 + 9)E[X;]. The 95th percentile requirement suggests that 9 > O. This extra amount, 9E[X;], is

the security loading and 9 is the relative security loading. Calculate 9. .


The criterion for e IS Pr(S :::; (l + S)E[S]) 0.95 where 5 X, + X2 + ... +

Xuoo· This probability statement is equivalent to

P [5 - E[5] <: eE[5] ] = 095

r v'Var(S) - v'Var(S) "

Following the discussion of the central limit theorem in Section 2.4, we approximate the distribution of (5 - E[5]) I v'Var(5) by the standard normal distribution and use its 95th percentile to obtain


It remains to calculate the mean and variance of 5 and to calculate e by this

For the four classes of insured individuals, we have the results given below.

Mean Variance
k qk bk bkql; b~qk(l - qk) nk
1 0.02 1 0.02 0.0196 500
2 0.02 2 0.04 0.0784 500
4 0.10 2 0.20 0.3600 500 l.~n(] 4

E[S] = 2: E[X;l = 2: l1,b,Qk = 160

, I '~I


1.800 4

Var(5) - L Var(X) - 2: Il,b;qr,(1 Gk) - 256

i-I k I

Thus, the relative security loading is

Chapter 2 Individual Risk Models for a Short Term


The policyholders of an automobile insurance company fall into two classes.

Distribution of Claim Amount, Bk,

Number Claim
Class in Class Probability
k n A L
1 500 0.10 1 2.5
2 2 000 0.05 2 5.0 A truncated exponential distribution is defined by the d.f.

{O x < 0

F(x) =_ ~ - e= 0 < x < L

X 2: L.

This is a mixed distribution with p.d.f. J(x) - '!I.e AI, 0 < X < L, and a probabihty mass e-AL at L. A graph of the d.f. appears in Figure 2.5.1.

Truncated Exponential Distribution F(x)

______ ~L_ ~~------~ X



Again, the probability that total claims exceed the amount collected from policyholders is 0.05. We assume that the relative security loading, e, is the same for the two classes. Calculate e.

This example is much like the previous one. It differs in that the claim amounts

are random variables. First, we obtain formulas for the moments of the truncated exponential distribution in preparation for applying (2.2.25) and (2.2.26):

Section 2.5 Applications to Insurance


e XL

1 - 2X.Le-AL - e-2AL

a2 = E[B2JI = 1] - (E[BJI = I]? = 2 •

Using the parameter values given and applying (2.2.25) and (2.2.26), we obtain the

following results.


Variance fJ.~qk(1 - qk) + (J"~qk

1 2

0.10 0.05

0.9179 0.5000

0.5828 0.2498

0.09179 0.02500

0.13411 0.02436

500 2 000

Then S, the sum of the claims, has moments

E[S] = 500 (0.09179) + 2,000 (0.02500) = 95.89,

Yar(S) = 500 (0.13411) + 2,000 (0.02436) = 115.78.

The criterion for e is the same as in Example 2.5.1,

Pr(S :oS (1 +e)E[S]) = 0.95.

Again by the normal approximation,

eE[S] = 1.645 \/Yar(S)






A life insurance company covers 16,000 lives for l-year term life insurance in amounts shown below.

Benefit Amount

Number Covered

20 000


50 000


The probability of a claim q for each of the 16,000 lives, assumed to be mutually independent, is 0.02. The company wants to set a retention limit. For each life, the retention limit is the amount below which this (the ceding) company will retain

Chapter 2 Individual Risk Models for a Short Term


the insurance and above which it will purchase reinsurance coverage from another (the reinsuring) company. For example, assume the retention limit is 20,000. The company will retain up to 20,000 on each life and purchase reinsurance for the difference between the benefit amount and 20,000 for each of the 4,500 individuals with benefit amounts in excess of 20,000. As a decision criterion, the company wants to minimize the probability that retained claims plus the amount that it pays

for reinsurance will exceed 8,250,000. Reinsurance is avaIlable at a cost of 0.025 per unit of coverage (i.e .. at 125% of the expected claim amount per unit, 0.02). We will

consider the block of business as closed. New policies sold during the year are not to ellter lItis decision process. Calculate the retention limit that minimizes the prob

ability that the company's retained claims plus cost of reinsurance exceeds 8,250,000.

Partial Solution:

First, do all calculations in benefit units of 10,000. As an illustrative step, let 5 be the amount of retained claims paid when the retention limit is 2 (20,000). Our

portfolio of retained business is given by

Retained Amount

Number Covered




8 000



E[S] = 2: /1kbkqk = 8,000 (1)(0.02) + 8,000 (2)(0.02) = 480

k -1



Var(S) = 2: ll,bt1h(1 - qk)

k 1

= 8,000 (1)(0.02)(0.98) + 8,000 (4)(0.02)(0.98) = 784.

In addition to the retained claims, 5, there is the cost of reinsurance premiums. The total coverage in the plan is

8,000 (1) + 3,500 (2) + 2,500 (3) + 1,500 (5) + 500 (10) = 35,000.

The retained coverage for tlte plan is

8,000 (1) + 8,000 (2) 24,000

Therefore, the total amount reinsured is 35,000 - 24,000 = 11,000 and the reinsur-

ance cost is 11,000(0.025) = 275. Thus, at retention limit 2, the retained claims plus reinsurance cost is 5 + 275. The decision criterion is based on the probability that this total cost will exceed 825,


Section 2. 5 Applications to Insurance

Pr(S + 275 > 825) = Pr(S > 550)

= Pr [S - E[5] > 550 - El5]]

VVar(5) VVar(S)

= Pr [5 - E[5] > 2.5].


Using the normal distribution, this is approximately 0.0062. The solution is completed il I Exercises 2.13 alld 2.14. T

In SectIon 1.5 stop-loss insurance, which is available as a reinsurance coverage,

was discussed. The expected value of the claims paid under the stop-loss reinsurance coverage can be approximated by using the normal distribution as the distribution of total claims.

Let total claims, X, have a normal distribution with mean j.L and variance a2 and let d be the deductible of the stop-loss insurance. Then, by (1.5.2A), the expected

claims equal

(2.5.1 )

Changing the variable of integration to z = (x - u) I a and defining 13 by d = j.L +

l3a, we obtain the following general expression for the expected value of stop-loss claims under a normal distribution assumption:

E[IAX)] = a {exP(-1302_/2) - 13[1 - <fJ(I3)]} (27T) .0

where <fJ(x) is the distribution function for the standard normal distribution.



Consider the portfolio of insurance contracts in Example 2.5.3. Calculate the expected value of the clailils provided by a stop-loss reinsurance coverage where

a. There is no individual reinsurance and the deductible amount is 7,500,000

b. There is a retention amount of 20,000 on individual policies and the deductible amount on the business retained is 5,300,000.


a. 'Nit}, no individual reinsurance and the use of 10,000 as the unit,

E[S] = 0.02[8000(1) + 3.500(2) + 2.500(3) + 1.500(5) + 500(10)] = 700


Var(5) = (0.02)(0.98)[8,000(1) + 3,500(4) + 2,500(9) + 1,500(25) + 500(100)] = 2,587.2

Chapter 2 Indi~'idual Risk Models for a Short Term



a(S) = 50.86.

Then, with

~ = (d - f.l) = (750 - 700) = 0.983

a 50.86

the application of (2.5 2) gives liS

P - 50.86[0.24608 - (0.983)(0.16280)] - 4.377.

This is equivalent to 43,770 in the example as posed.

b. In Example 2.5.3 we determmed the mean and the variance of the aggregate claims, after imposing a 20,000 retention limit per individual, to be 480 and 784, respectively, in units of 10,000. Thus a(S) = 28.

Then, with

d - f.l

530 - 480


the application of (2 5 2) gives us

P - 28[0.08100 (1.786)(0.03707)] - 0.414

This is equivalent to 4,140 in the example as posed.

2.6 Notes and References

The basis of the material in Sections 2.2, 2.3, and 2.4 can be found in a number of post-calculus probability and statistics texts. Mood et a!. (1974) prove the theorems given in (2.2.10) and (2.2.11). They also provide an extensive discllssion of

properties of the moment generating function. For a discussion of the advanced mathematical methods for deriving the distribution function conespollds to a

given moment generating function, see Bellman et al. (1966). Methods are also

aVdilable to obtain the p.f. of a dIscrete dIstnbution from its probability generating function; see Kornya (1983).

DeGroot (1986) provides a discussion of several conditions under which the centrallimit theorem holds. Kendall and Stuart (1977) give material on normal power expansions that may be viewed as modifications of the normal approximation to improve numerical results. Bowers (1967) also describes the use of normal power

expansions and gives an application to approximate the distribution of present values for an annuity portfolio

Section 2.6 Notes and References


Section 2.2

2.1. Use (2.2.3) and (2.2.4) to obtain the mean and variance of the claim random variable X where q = 0.05 and the claim amount is fixed at 10.

2.2. Obtain the mean and variance of the claim random variable X where q = 0.05

and the claim aIlloullt random vanable B IS umformly distributed between 0 and 20 .

. L

2:3. Let X be the number of heads observed in five tosses of a true coin. Then, X true dice are thrown. Let Y be the sum of the numbers showing on the dice. Determine the mean and variance of Y. [Hint: Apply (2.2.10) and (2.2.11).]

2.4. Let X be the number showing when one true die is thrown. Let Y be the number of heads obtained when X true coins are then tossed. Calculate

E[Y] and Var(Y).

2.5. Let X be the number obtained when one true die is tossed. Let Y be the sum

of the numbers obtained when X true dice are then thrown. Calculate E[Y] and Var Y .

2.6. The probability of a fire in a certain structure in a given time period is 0.02.

If a fire occurs, the damage to the structure is uniformly distributed over the interval (0, a) where a is its total value. Calculate the mean and variance of fire damage to the structure within the time period.

2.7. Independent random variables Xk for four lives have the discrete probability functions given below.

x PriX! = .r) Pr(Xz = x) Pr(X3 = x) Pr(X .. = xl
o 0.6 0.7 0.6 0.9
1 0.0 0.2 0.0 0.0
2 0.3 0.1 0.0 0.0
3 0.0 0.0 0.4 0.0
4 0.1 0.0 Use a convolution process on the non-negative integer values of x to obtain

Fj(x) for x 0, 1, 2, I 13 where 5 XI I X2 I X3 + X •.

2.S. Let Xl for i = 1, 2, 3 be independent and identically distributed with the d.f.

{o x < 0

F(x) = 'IX 0 ::s x < 1

X 2:: 1.

Chapter 2 Individual Risk Models for a Short Term

Let S = XI + X2 + X3·

a. Show that Fs(x) is given by

s .





,/ .> " . .5

~~ J -'




X 2:: 3.

b. Show that E[S] = 1.5 and Var(S) = 0.25.

c. Evaluate the following probabilities using the d.f. of part (a):

2.9. Find the mean and variance of the inverse Gaussian distribution by using its m g f as given in Example 2.3.5.

Section 2.4

2.10. Calculate the mean and variance of X and Y in Example 2.3.2. Use a normal distribution to approximate Pr(X + Y > 4). Compare this with the exact answer.

2.11. a. Use the central limit theorem to calculate b, c, and d, for given 11, in the


Pr (* XI 2:: nj.L + i1v;~a) ~ C + bctJ(d)

where the X/ s are independent and identically distributed with mean f.l and variance a" and ctJ(z) is the d.f. of the standard normal distribution.

b. Evaluate the probabilities in Exercise 2.8(c) by use of the normal approximation developed in part (a).

2.12. A random variable U has m.g.f.

a. Use the m.g.f. to calculate the mean and variance of U.

b. Use a normal approximation to calculate points YOlJ5 and YOOI such that Pr(U > Y.) = E.

Note the random variable U has a gamma distribution with parameters a = 9 and ~ = 1/2. Gamma distributions with a = n /2 and ~ = 1/2 are chi-square distributions with 11 degrees of freedom. Thus U has a chi-square distribution with 18 degrees of freedom. From tables of d.f.'s of chi-square distributions, we obtain Yo os = 28.869 and Yom = 34.805.

2.13. Calculate the probability that the total cost in Example 2.5.3 will exceed 8,250,000 if the retention limit is

a. 30,000 b. 50,000.

2.14. Calculate the retention limit that minimizes the probability of the total cost in Example 2.5.3 exceeding 8,250,000. Assume that the limit is between 30,000 and 50,000.

2.15. A fire insurance company covers 160 structures against fire damage up to an amount stated in the contract. The numbers of contracts at the different con-

tract amounts are given below.

Contract Amount

Number of Contracts

10 000 20 000 30 000 50 000

100 000

80 35 25 15


Assume that for each of the structures, the probability of one claim jrlTithin a

year is 0.04, and the probability of more than one claim is O. Assume that

fires in the structures are mutually independent events. Furthermore, assume that the conditional distribution of the claim size, given that a claim has oc-

curred. is uniformly distributed over the interval from 0 to the contract amount. Let N be the number of claims and let 5 be the amount of claims in a J-vear period.

a. Calculate the mean and variance of N.

b. Calculate the mean and variance of S.

c. What relative security loading, 8, should be used so the company can collect an amount equal to the 99th percentile of the distribution of total

claims? (Use a normal approximation.)

2.16. Consider a portfolio of 32 policies. For each policy, the probability q of a claim

is 1 76 and B, the benefit amount given that there is a claIm, has p.d.f.

fll/) = {2(1 - y) 0 < y < 1

. . 0 elsewhere.

Let 5 be the total claims for the portfolio. Using a normal approximation, estimate Pr(S > 4).

Chapter 2 Individual Risk Models fOl a 511011 Telm


,,-=- .C-~
3 ,


3.1 Introduction

Chapter 1 was dedicated to showing how insurance can increase the expected utility of individuals facing landom losses. In Clraptel 2 simple models for single-

period insurance policies were developed. The foundations of these models were

Bernoulli random variables associated with the occurrence or nonoccurrence of a loss. The occurrence of a loss, in some examples, resulted in a second random process generating the amount of the loss. Chapters 4 through 8 deal primarily with models for insurance systems designed to manage random losses where the randomness is related to how long an individual will survive. In these chapters the time-until-death random variable, T(x), is the basic building block. This chapter develops a set of ideas for describing and using the distribution of time-until-death

and the distribution of the corresponding age-at-death, X.

We show how a distribution of the age-at-death random variable can be sum-

rnarized by a life table. Such tables are useful in many fields of science. Consequently a profusion of notation and nomenclature has developed among the various professions using life tables. For example, engineers use life tables to study the reliability of complex mechanical and electronic systems. Biostatisticians use life tables to compare the effectiveness of alternative treatments of serious diseases. Demographers LIse life tables as tools in population projections. In this text, life tables are used to build models for insurance systems designed to assist individuals

facing uncertainty about the times of their deaths. This application determines the viewpoint adopted. However, 'Nhen it provides a bridge to other disciplines, notes

relating the discussion to alternative applications of life tables are added.

A life table is an indispensable component of many models in actuarial science.

In fact, some scholars fix the date of the beginning of actuarial science as 1693. In that year, Edmund Halley published" An Estimate of the Degrees of the Mortality of Mankind, Drawn from Various Tables of Births and Funerals at the City of

Chapter 3 Survi,'ai Distributions and Life Tables


\._. /1'.1 -


mains of interest because of its surprisingly modern notation and ideas.

Breslau-" Tlie life table, called the Breslau Table, contained in Halley's paper re-


Probability for the Age-at-Death

In this section we formulate the uncertainty of age-at-death in probability



The Survival Function

Let us consider a newborn child. This newborn's age-at-death, X, is a continuoustype random variable. Let Fx(x) denote the distribution function (d.f.) of X,

Fx(x) = Pr(X ::; x)

X 2: 0,

(3.2.1 )

and set

Fx(x) Pr(X > x)

X 2: O.


called the survival function (s.f.). For any positive x, SIX) is the probability a new-

We always assume that Fx (0) 0, wlIich implies 5(0) 1. The function SIX) is

born wIll attain age x. The distribution of X can be defined by specifying either the function F,,(x) or the function sex). Within actuarial science and demography, the

survival function has traditionally been used as a starting point for further developments. Within probability and statistics, the d.f. usually plays this role. However, from the properties of the d.f., we can deduce corresponding properties of the survival function.

Using the laws of probability, we can make probability statements about the ageat-death in terms of either the survival functioll 01 tlie distdbutiOIl functton: For

example, the probability that a newborn dies between ages x and z (x < z) is

= s(x) - s(z).


Time-until-Death for a Person Age x

The conditional probability that a newborn will die between the ages x and z, given survival to age x, is

Pr(Y < X < zl X > x)

sex) s(z)



The symbol (x) is used to denote a life-age-x. The future lifetime of (x), X - x, is denoted by T(x). '\lX-)-'" i',- X


Section 3.2 Probability for the Age-at-Death

Within actuarial science, it is frequently necessary to make probability statements

about T(x). For this purpose, and to promote research and communication, a set of symbols, part of the International Actuarial Notation, was originally adopted by the 1898 International Actuarial Congress. Symbols for common actuarial functions and principles to guide the adoption of new symbols were established. This system has been subject to constant review and is revised or extended as necessary by the International Actuarial Association's Permanent Committee on Notation. These no-

tahona! conventions are followed In this book whenever possible.

These symbols differ from those used for probability notation, and the reader may be unfamiliar '.vith them. For example, a single-variate 6mction that would be

written q(x) in probability notation is written q, in this system. Likewise, a multivariate function is written in actuarial notation using combinations of subscripts, superscripts, and other symbols. The general rules for defining a function in actuarial notation are given in Appendix 4. The reader may want to study these forms before continuing the discussion of the future-lifetime random variable.

10 make probabilIty statements about T(x), we use the notations

ICf, Pr[T(x) -e-: t]

t "> 0,




Iq, Pr[T(x) > t]

t :> 0

The symbol N, can be interpreted as the probability that (x) will die within t years: \

t f_

~ u.r 1ft\/", I .f-v. w:t:r(:~ \ +v i I \

that is, H, is the d.f. of T(x). On the other hand, IV, can be interpreted as the prob- i ability that (x) will attain age x + t; that is, p, is the s.f. for (x). In the special case of a life-age-O, we have T(O) = X and


,Po = sex) x 2: O. Ci-£ is o ()rrj ~urt1\ie (3.2.6)

If t = 1, convention permits us t:o~{{de prefix in the symbols defined in (3.2.4)

q, = Pr[(x) will die within 1 year],

_1 __ j~l~. )< t t*'cl

].I, = Pr[(x) will attain age x + 1].

There is a special symbol for the more general event that (x) will survive t years and die within the following u years; that is, (x) will die between ages x + t and x + t + u. This special symbol is given by

IllIq, = Pr[t < T(x) :s t + u]


As before, if u = 1, the prefix is deleted in tlllq" and we have Ilqx'

At this point it appears there are two expressions for the probability that (x) will die between ages x and x + u. Formula (3.2.7) with t = 0 is one such expression; (3.2.3) with z = x + u is a second expression. Are these two probabilities different? Formula (3.2.3) can be interpreted as the conditional probability that a newborn

Chapter 3 Survival Distributions and Life Tables


WIll dIe between ages x and z x + II, given survival to age x. The only information on the newborn, now at age x, is its survival to that age. Hence, the probability statement is based on a conditional distribution of survival for newborns,

On the other hand, (3.2.7) with t = 0 defines a probability that a life observed at age x will die between ages x and x + u. The observation on the life at age x might include information other than simply survival. Such information might be that the

life has just passed a physical examination for insurance, or it might be that the life has commenced treatmel1t fOI a serious Hitless. Life tables for situations where

the observation of a life at age x implies more than simply survival of a newborn

to age x are dIscussed In Section 3.8, where additional notation for those life tables is introduced. We will continue development of the theory without further reference to the distinction between (3.2.3) and (3.2.7), and~'H'e aciSlJu1ethatuntiIJhm

1_t~S!~Qn.o5se;,vatTcilloL5:11r~Tval ~Tige -""--'ym-Yie1(:i. the-s~p1e,~n:dlfi_Qna1 ffi5trFmIDn ~!!!..!)£~_thehypot1Tg:SUf1hClta ne,y):)S.?:n:d\as survJ\'e¢ltoag_e:~; that is,

- ,- IV) sex + t)
.\P(l S x
1 :, (3.2.9)
Iq\ = -
sex) Under this approach, (3.2.7), and its many special cases, can be expressed as

sex + t) - sex + t + 1I)

tluq.\ ==


= [sex + t)] [sex +


t) - sex + t + sex + t)




A dIscrete random vanable associated with the future lifetime is the number of future years completed by (x) prior to death. It is called the curtate-future-lifetime of (x) and is denoted by K(x). Because K(x) is the greatest integer in T(x), its p.f. is

Pr[K(x) = k] = Pr[k :::; T(x) < k + 1]

= Pr[k < T(x) :::; k + 1]

k- 0, 1, 2, ....


The switching of inequalities is possible since, under our assumption that T(x) is a

continuous-type random variable, Pr[T(x) - k] - Pr[T(x) - k + 1] - O. Expression (3.2.11) is a special case of (3.2.7) where u = 1 and k is a non-negative integer. From (3.2.11) we can see that the d.f. of K(x) is the step function


FK(l) (Y) = ~ lJ!q\ = k+1Q"


_II 2:: 0 and k is the greatest integer in y.


Section 3.2 Probabilitv for the Age-at-Death

It often follows from the context that T(x) is the future lifetime of (x), in which case we may write T instead of T(x). Likewise, we may write K instead of K(x).


Force of Mortality

Formula (3.2.3) expresses, in terms of the d.f. and in terms of the survival func-

tion, the conditional probability that (0) will die between ages x and z, given survival to x With z - x held constant, say, at c, then considered as a function of x,

this conditional probability describes the distribution of the probability of death in the near future (between time 0 and c) for a life of attained age x. An analogue of

this function for instantaneous death can be obtained by using the density of probability of death at attained age x, that is, using (3.2.3) with z = x + 1x,

P (X .~ IX ) Fx(x + Ix) - Fx(x)

r x < :::; x + uX > x = -'-'--'-----'-----'-'--'_:_

1 - Fx(x)

x (;r)..lx

In this expression F~(x) = /x(x) is the p.d.f. of the continuous age-at-death random variable. The function


1 - Fx(x)

in (3.2.12) has a conditional probability density interpretation. For each age x, it gives the value- of the conditional p.d.f. of X at exact age x, given survival to that age, and is denoted by j.1(x).

We have


The properties of /x(x) and 1 - Fx(x) imply that j.1(x) 2: O.


In actuarial science and demography j.1(x) is called the force of mortality. In reliability theory, the study of the survival probabilities of manufactured parts and systems, jJ (x) is called the failure rate or hazard rate or, more fully, the hazard

rate function.

As is true for the s.f., the force of mortality can be used to specify the distribution

of X. To obtain this result, we start with (3.2.13), change x to y, and rearrange to obtain

- j.1(y) dy = d log s(y).

Chapter 3 Survival Distributions and Life Tables


Integrating this expression from x to x + II, we have

JX~II [sex + II)]

- /-ley) dy = log ,

x 50)

= log "p."

and on taking exponentials obtain

,Px exp[ I:+II /-l(Y) dy].


Sometimes it is convenient to rewrite (3.2.14), with", Y x, as

s. exp[ r~ Il-(x + s) ds].


In particular, we will change the notation to conform with that used in (3.2.6) by setting the age already lived to 0 and denoting the time of survival by .r. We then have

.,Po = sex) = exp[ - n /-l(5) ds].


In addition,

Fx(x) 1 sex) 1 exp] n /-l(s) ds]



Let FT(Jt) and !rc,)(t) denote, respectively, the d.f. and p.d.f. of T(xj, the future lifetime of (x). From (3.2.4) we note that FT,,)(t) = tq,; therefore,


jnxP) = dt tq,

:t [I S(:(:) t)]

_ sex + t) [ sex)

s'(x + t)] sex + t)

= tPx /-l(x + t)

t 2: O.


Thus tPx /-l(x + t) dt is the probability that (x) dies between t and t + dt, and LX tPx /-l(X + t) dt = 1

for integrating over all positive probability density).

where the upper limit on the integial is written as posiflve mfInIty (an abbreviation

It follows from (3.2.19) that

d d

dt (1 - fPJ = - dt tP, = tV, /-l(x + t).


This equivalent form is useful in several developments in actuarial mathematics.

Section 3.2 Probability for the Age-at-Death


we have

lim IIP.\ = 0,


lim (-log "Px) = x;

that is,

The developments of this section are summarized in Table 3.2.1.

The lower half of Table 3.2.1 summarizes some of the relationships among functions of general probability theory and those specific to age-at-death app!i<::.ations~ There are many other examples where age-at-death questions can be formed in the more general probability setting. The following will illu5tIate this point.

Probability Theory Funchons for Age-at-Death, X


x<o x = 0 X2:0 lim


d.f. Fx(x)

Survival Function sex)

Force of

p.d.f. /x(x)

Mortality J.l(x)


x < 0 ix(x) = 0 x = 0 undefined x > 0 fx(x) 2: 0

lim j 0 fx ex) dt


J.L(x) = 0 undefined J.L(x) 2: 0

I J~ J.L(x)dx



Fx(x) = 0 Fx(O) = 0

sex) = 1 5(0) = 1


in Terms of

nondecreasing FxCx) 1


5(3.) 0

Fx(x) sex) fx(X) J.L(x)


1 - sex) Iofx(u)du

1- exp[-IoJ.L(t)dt]

1 - Fx(x) sex)

I: fx(U)du

exp] - g J.L(t) dt]

Fx(x) -s'(x) fx(x)

J.L(x)exp[ - 10 J.L(t) dt]

F~(x)/ [1 - Fx(x)] -s'(x) / sex)

fx(X) / I: fx(u)du J.L(X)

1;:£1,,1,15'·_1================================-----------If A refers to the complement of the event A within the sample space and

Pr(A) #= 0, the foIIowmg expresses an identity in probability theory:

PI (A U B) Pr(A) + Pr(A) Pr(BIA).

Rewrite this identity in actuarial notation for the events A = [T(x) ::; t] and B [t < T(x) ::; 1], ° < t < 1.

Chapter 3 Survival Distributions and lite Tables 57

Pr(A U B) becomes Pr[T(x):s:: 1] = q" Pr(A) is tq" and Pr(BIA) is llq,.t; hence

3.3 Life Tables

A published life table usually contains tabulations, by individual ages, of the basic functions q" Iv dr' CIIld, possibly, additional derIved tunctlOns. Before pre-

related to the probability functions discussed in the preceding section.

senting such a table, we consider an interpretation of these functions that is directly


Relation of Life Table Functions to ttle Survival Function

In (3.2.9) we expressed the conditional probability that (x) will die within t years by


\ sex + t)


and. in particular, we have


We now consider a group of 10 newborns, 10 = 100,000, for instance. Each newborn's age-at-death has a distribution specified by s.f. sex). In addition, we let .~'(x) denote the group's number of survivors to age x. We index these lives by j = I, 2, ... , 10 and observe that


X(x) 2: I,

, 1

where I, is an indicator for the survival of life i that is,

I = {I if life j survives to age x

, 0 otherwise.

Since E[Ij] = 5(X),


E[,I:'(x)] = 2: E[I;J = 10 seX).


We denote E[.I:'(x)] by I,; that is, I, represents the expected number of survivors to age x from the 10 newborns, and "\Ie have

I 5 X .

Moreover, under the assumption that the indicators I, are mutually independent, ,\:'(x) has a binomial distribution with parameters J1 = 10 and p = sex). Note, however, that (3.3.1) does not require the independence assumption.


Section 3.3 Life Tables

In a similar fashion, ,,'J~, denotes the number of deaths between ages x and x + II from among the initial '0 lives. We denote E[,,'J\] by lId,. Since a newborn has probability s(x) - s(x + 11) of death between ages x and x + II we can, by an argument similar to that for Iv express

,A, = E[,/J\] = lo[s(x) - s(x + 11)]

When 11 = 1. we omit the prefixes on ,,'J\ and "dr·

From (3.3.1), we see that

_ ! c!!2

I, dx

1 ds(x)



s(x) dx


-dI, = I, f..L(x) ax.


the factor I, f..L(x) in (3.3.4) can be interpreted as the expected density of deaths in

the age interval (x, x + dx). We note further that

I, - 10 exp[ Jl\ f..L(y) dy],

(3.3.:J ) (3.3.6) (3.3.7)

1\+11 = I, exp] - J:+II f..L(y) dy],

I, - 11+11 = J:+II III f..L(y) dy.

For convenience of reference, wecall.this concept of/l~-~ewborn~, ~<::h with

~':1rvival fu~~00~s~x),a random 5u-rvivorship group. i ,- .:"


Life Table Example

In "Life Table for the Total Population: United States, 1979-81" (Table 3.3.1), the functions tq" I" and .d, are presented with 10 = 100,000. Except for the first year of life, the value of t in the tabulated functions tq, and l, is 1. The other functions appearing in the table are discussed in Section 3.5.

The 1979-81 Ll.S, Life Table was not constructed by observing 100,000 newborns until the last survivor died Instead, jt was based on estimates of probabilities of

death, given survival to various ages, derived from the experience of the entire U .5. population ill the yeals aIOulld the 1980 census. III uSillg the random surA

vorship group concept with this table, we must make the assumption that the

probabIlities derived from the table will be appropriate for the lIfetimes of those ~~~ b~~ong~o the survivorship group.

Chapter 3 Survival Distributions and Life Tables


Life Table for the Total Population: United States, 1979-81
(1) (2) (3) (4) (5) (6) (7)
Proportion Remaining
Dying Lifetime
Proportion of Stationary Population" Average
Persons Alive Number of
at Beginning Of 100.000 Born Alive Years LIved Years of Life
Age Interval of Age Number in This Remaining
Period of Interval Living at Number Years Lived and All at Beginning
Life between Dying during Beginning of Dying during in the Age Subsequent of Age
Two Ages Interval Age Interval Age Interval Interval Age Intervals Interval
x to x + t 1, tdx ,Lx r, 0
Ilx e,
0-1 0.00463 100000 463 273 7387758 73.88
1-7 0.00246 99537 245 1635 7387485 74.22
7-28 0.00139 99292 138 5708 7385850 74.38
28 365 0.00418 99154 414 91357 .{ 7380142 74.43
0.01260 100000 1260 98973 7387758 73.88
2 0.00093 98740 92 98694 7288785 73.82
2 3 0.00065 98648 64 98617 7190091 72.89
3 4 0.00050 98584 49 98560 7 091 474 7193
4 5 0.00040 98535 40 98515 6992914 70.97
5-6 0.00037 98495 36 98477 6894399 70.00 \
6-7 0.00033 98459 33 98442 6795922 69.02
7-8 0.00030 98426 30 98412 6697480 68.05
8-9 0.00027 98396 26 98383 6599068 67.07
9-10 0.00023 98370 23 98358 6500685 66.08
. ~,
1011 0.00020 98347 19 \, 98338 6402327 65.10
11-12 0.00019 98328 19 98319 6303989 64.11
12 13 0.00025 98309 24 98297 6205670 63.12
13-14 0.00037 98285 37 98266 6107373 62.14
14 15 0.00053 98248 52 98222 6009107 61.16
15-16 0.00069 98196 67 98163 5910885 60.19
16-17 0.00083 98129 82 98087 5812722 59.24
17-18 0.00095 98047 94 98000 5714635 58.28
18-19 0.00105 97953 102 97902 5616635 57.34
19-20 0.00112 97851 110 97796 5518733 56.40
20-21 0.00120 .. 9774~ 118 97682 5420937 55.46
21 22 0.00127 9762 f24 97561 5323255 54.53
22-23 0.00132 97499 129 97435 5225694 53.60
23 24 000134 97370 130 97306 5128259 ::>2.67
24-25 0.00133 97240 130 97175 5030953 51.74
::> 6 0.00132 97110 )2~ 97046 4933778 50.81
26-27 0.00131 96982 126' 96919 4836732 49.87
27-28 0.00130 96856 126 96793 4739813 48.94
28-29 0.00130 96730 126 96667 4643020 48.00
29-30 0.00131 96604 127 96541 4546353 47.06
'Stationary population is a demographic concept treated in Chapter 19. 60

ection 3.3 Life Tables

Life Table for the Total Population: United States, 1979-81
(1) (2) (3) (4) (5) (6) (7)
Proportion Average
Dying Remaining
Proportion of Lifetime
Persons Stationary Population" AveIage
Alive Number of
at Beginmng 9f H)9,999 Bem Ali .. -e Years LIved Years of LHe
Age Interval of Age Number in This Remaining
PerIod of Interval Living at Number Years Lived and All at Beginning
Life between Dying during Beginning of Dying during in the Age Subsequent of Age
Two Ages Interval Age Interval Age Interval Interval Age Intervals Interval
x to x + t t, tdx tLx T,. 0
Ilx ex
30-31 0.00133 96 477 127 96414 4449812 46.12
31-32 0.00134 96350 130 96284 4353398 45.18
32-33 0.00137 96 220 132 96155 4257114 44.24
33 34 0.00142 96088 137 96019 1160959 13.30
34-35 0.00150 95951 143 95880 4064 940 42.36
35-36 0.00159 95808 153 95731 3969060 41.43
36 37 000170 95655 163 95574 3873329 40.49
37-38 0.00183 95492 175 95 404 3777755 39.56
38-39 000197 95317 188 95224 3682351 38.63
39-40 0.00213 95129 203 95027 3587127 37.71
40-41 0.00232 94926 220 94817 3492100 36.79
41-42 0.00254 94706 241 94585 3397283 35.87
42-43 0.00279 94465 264 94334 3302698 34.96
43-44 0.00306 94201 288 94057 3208364 34.06
44-45 0.00335 93913 314 93756 3114307 33.16
45 46 0.00366 93599 343 93427 3020551 32.27
46-47 0.00401 93256 374 93069 2927124 31.39
47 48 0.00442 92882 410 92677 2834055 30.51
48-49 0.00488 92472 451 92246 2741378 29.65
49 50 0.00538 92021 495 91773 2649132 2879
50-51 0.00589 91526 540 91256 2557359 27.94
51-52 0.00642 90986 584 90695 2466103 27.10
52-53 0.00699 90402 631 90086 2375408 26.28
53-54 0.00761 89771 684 89430 2285322 25.46
54-55 0.00830 89087 739 88717 2195892 24.65
55-56 0.00902 88348 797 87950 2107175 23.85
56 57 0.00978 875::>1 856 87122 2019225 23.06
57-58 0.01059 86695 919 86236 1932103 22.29
58 59 0.01151 85776 987 85283 1845867 21.52
59-60 0.01254 84789 1063 84258 1760584 20.76
60-61 0.01368 83726 1145 83153 1676326 20.02
61-62 0.01493 82581 1233 81965 1593173 19.29
62-63 0.01628 81348 1324 80686 1511 208 18.58
63-64 0.01767 80024 1415 79316 1430522 17.88
64-65 0.01911 78609 1502 77859 1351206 17.19
'Stationary population is a demographic concept treated in Chapter 19. Chapter 3 Survival Distributions and Life Tables


Life Table for the Total Population: United States, 1979-81
(1) (2) (3) (4) (5) (6) (7)
Proportion Average
Dying Remaining
Proportion of Lifetime
Persons StationaIY Population* Average
Alive Number of
at Beginning Of 100,000 Born Alive Years Lived Years of Life
Age Interval of Age Number In This RemaInIng
Period of Interval Living at Number Years Lived and All at Beginning
I He between Dying during Beginning of Dying during in the Age Subsequent of Age
Two Ages Interval Age Interval Age Interval Interval Age Intervals Interval
x to x + t t, td, (LX Tx 0
tqx ex
65-66 0.02059 77107 1587 76314 1273347 16.51
66-67 0.02216 75520 1674 74683 1 197033 15.85
67-68 0.02389 73846 1764 72964 1 122350 15.20
68-69 0.02585 72 082 1864 71150 1049386 14.56
69 70 o 0?806 70218 1970 69233 978236 13.93
70 71 0.03052 68248 2083 67206 909003 13.32
71-72 0.03315 66165 2193 65069 841797 12.72
72 73 0.03593 63972 2299 62823 776728 12.11
73-74 0.03882 61673 2394 60476 713 905 11.58
74 75 0.04184 59279 2480 58039 653429 11 02
75-76 0.04507 56799 2560 55520 595390 10.48
76-77 0.04867 54239 2640 52919 539870 9.95
77-78 0.05274 51599 2721 50238 486951 9.44
78-79 0.05742 48878 2807 47475 436713 8.93
79-80 0.06277 46071 2891 44626 389238 8.45
80-81 0.06882 43180 2972 41694 344612 7.98
81 82 007552 10 208 3036 38689 302918 7.53 ......
82-83 0.08278 37172 3077 35634 264229 7.11
83 84 0.09041 34095 3 083 32553 228595 6.70
84-85 0.09842 31012 3052 29486 196042 6.32
85-86 0.10725 27960 2999 26461 166556 5.96
86-87 0.11712 24961 2923 23500 140095 5.61
87-88 0.12717 22038 2803 20636 116595 5.29
88-89 0.13708 19235 2637 17917 95959 4.99
89-90 0.14728 16598 2444 15376 78042 4.70
90-91 0.15868 , 14154 . 2246 13 031 62666 4.43
91-92 0.17169 11908 2045 10886 49635 4.17
92 93 0.18570 9863 1831 8948 38749 3.93
93-94 0.20023 8032 1608 7228 29801 3.71
94 95 0.21495 6424 1381 5733 22573 3.51
95 96 022976 5043 1159 4463 16840 3.34
96-97 0.24338 3884 945 3412 12377 3.19
97-98 0.25637 2939 754 2562 8965 3.05
98-99 0.26868 2185 587 1892 6403 2.93
99-100 0.28030 1598 448 1374 4511 2.82
'Stationary population is a demographic concept treated in Chapter 19. 62

Section 33 Life Tables

Life Table for the Total Population: United States, 1979-81
(1) (2) (3) (4) (5) (6) (7)
Proportion Average
Dying Remaining
Proportion of Lifetime
Persons Stationary Population* Average
AlIve Number of
at Beginning Of 100,000 Born Alive Years Lived Years of Life
A ge Interval of Age Number in This Remaining
Period of Interval Living at Number Years Lived and All at Beginning
Life between Dying dUling Beginning of Dying dUIing in the Age Subsequent of Age
Two Ages Interval Age Interval Age Interval Interval Age Intervals Interval
x to x + t tqx Ix A,· ~x Tx e;
100-101 0.29120 1 150 335 983 3137 2.73
101-102 0.30139 815 245 692 2154 2.64
102-103 0.31089 570 177 481 1462 2.57
103-104 0.31970 393 126 330 981 2.50
104 105 0.32786 267 R8 223 651 2.44
1()1) 106 0331)39 179 60 11)0 4')8 238
106-107 0.34233 119 41 99 278 2.33
107-108 0.34870 78 27 64 179 2.29
108-109 0.35453 51 18 42 115 2.24
109-110 0.35988 33 12 27 73 2.20
'Stationary population is a demographic concept treated in Chapter 19. Several observations about the 1979-81 Ll.S, Life Table are instructive.


1. Approximately 1% of a SUI vivors!tip group of lIeWbOlIlS would be expected to

die in the first year of life.

2. It would be expected that about 77°/r., of a group of newborns would survive to a e 65.

3. The maximum number of deaths within a group would be expected to occur between ages 83 and 84.

4. Forhuman lives, there have been few observations of age-at-death beyond 110.

Consequently, it is often assumed that there is an age w such that s(x) > 0, for x < w,and s(x) = 0 for x 2 w. The age w, if assumed, is called the limiting ~age.

"The limiting age for this table is not defined. It is clear that there is a positive probability of survival to age 110, but the table does not indicate the age w.

5. Local minimums in the expected number of deaths occur around ages 11 and

27 and a local maximum ,around age 24.

6. Although the values ofUhave been rounded to integers, there is no compelling

reason, according to (33.1), to do so.

A display such as Table 3.3.1 is the_conventional method.for describing the. dis-, tribution of age-at-death. Alternatively, an s.f. can be described in analytic form s-uch as s(x) = e=, c > 0, X 2 O. However, most studies of human mortality for

Chapter 3 Survival Distributions and Life Tables


insurance purposes use the representation s(x) - It 110' as illustrated in Table 3.3.1. Since 100,000 5(X) is displayed for only integer values of x, there is a need to interpolate in evaluating 5(X) for noninteger values. This is the subject of Section 3.6.

a. Live to 100

b. Di

On the basis of Table 3.3.1, evaluate the probability that (20) will

c. Die in the tenth decade of life.


a. 5(100) = 1100 = 1,150 = 0.0118

5(20) 1211 97,741

b. [5(20) - 5(70)] = 1 _ 170 = 1 _ 68,248 = 0.3017

[5(90) - 5(100)]



(14,154 - 1,150)




Insight into life table functions can be obtained by studying Figures 3.3.1, 3.3.2, Elnd 3.3.3. These are dra"vn to be represelltative of CUI lent human mortality and

are not taken directly from Table 3.3.1. J

Graph of f.l(x)






\. ---

x Age

11 111


Section 3.3 Life Tables

In Figure 3.3.1 note two features:

• The force of mortality is positive and the requirement

fx fJ-(x) dx = oc


appears satisfied. (See Table 3.2.1.)

• The force of mortality starts ant rather large and then drops to a minimum

around age 10.

Graph of Ix f.1(x)









x Age

Graph of Ix


0: 60
40 o




Chapter 3 Survival Distributions and Life Tables


In Figures 3.3.2 aud 3.3.3 note tlte following:

• The function l, !-L(x) is proportional to the p.d.f. of the age-at-death of a newborn. Since l, !-L(x) is the expected density of deaths at age x, under the random survivorship group idea, the graph of Ix !-L(x) is called the curve of deaths.

• There is a local minimum of i, !-L(x) at about age 10. The mode.of the distribution of deaths-=-t~~~_!_w~i_ch the _,maximum of the curve of deaths occur~-is around age 'SO.

• The function l, is proportional to the survival function s(x). It can also be in> terpreted as the expected number living at age x out of an illitial group of'

size 10,

• Local extreme points of t; !-L(x) correspond to points of inflection of l, since

d d ( d) d2

-d I, !-L(x) = -d - -d I, = - -d 7 I, .

x . x x . x- .

3.4 The Deterministic Survivorship Group

We proceed now to a second, and nonprobabilistic, interpretation of the life table.

This is rooted mathematically in the concept of decrement (negative growth) rates.

As such, it is related to growth-rate a lications in biolo and economi s.

eterministic in nature and leads to the concept of a d~te inistic survivorship group or cohort

A deterministic survivorship group, as represented by a life table, has the following characteristics:

• The group initially consists of 10 lives age O .

• The members of the group are subject, at each age of their lives, to effective annual rates of mortality (decrement) specified by the values of qx in the life

• The group is closed. No further entrants are allowed beyond the initial/p• The

only decreases come as a result of the effective annual rates of mortality decrement.

From these characteristics it follows that the progress of the group is determined by

I] = 10(1 - qo) = 10 - do,

12 = 1](1 - q]) = I] - d] = 10 - (do + d]),

.r ]

(3.4.1 )


Section 3.4 The Deterministic Survivorship Group

where I, IS the number of lives attaining age x in the survivorship group. This chain of equalities, generated by a value '(1 called the radix and a set of q, values, can be rewri tten as

1, I, 1 Px 1


There is an analogy between the deterministic survivorship group and the model for compound interest. Table 3.4.1 is designed to summarize some of this parallelism.

Related Concepts of the Mathematics of Compound Interest

and of Deterministic Survivorship Groups

Componnd Interest

Survivorship GlOup

A(f) = Size of fund at time t, time measured in years

I, = Size of group at age x, age lIIeasured in years

Effective annual rate of interest (increment)

Effective annual rate of mortality (decremen t)

A(f + 1) - A(f)




I, - 1'.1 I,

Effective ll-year rate of interest,

Effective ll-year rate of mortality,

sta] lillg a t time i

starting at age x

Force of interest at time t

Force of mortality at age x

8 = lim [AU + At) - A(t)]

1 .:It-(I A(t) At

1 dA(t)


A(t) dt

_ _!_ dl,

I, dx

'There is no universally accepted symbol for an effective n-year rate of interest

The headings of the Iqv I" and tdx columns in rable 3.3.1 refer to the deterministic survivorship group interpretation. j\lthough the mathematical foundations()f the

.randorn survivorship group and the deterministic survivorship group aredifferent,!1nctions q" I), and d,,_~ave the same mathematical properties_ and _suJ:,sequent analysis_. The random survivorship group concept has th~_advantage of allo_~ingfor the.full useof.probabjlity theory.The deterministic survivorship group

Chapter 3 Survival Distributions and Life Tables


an easy to apply but does not take account of random variation in the number of survivors,

3.5 Other Life Table Characteristics

In this section we derive expressions for some com _

e istri utions of T(x) and K(x) and introduce a general method for computing several of these characteristics




The expected value of T(x), denoted bye" is called the complete-expectationof-life. By definition and an integration by parts, we have

e, = E[T(x)] = (X / tP, JL(x + t) dt


t( tpJl~ + f tV, dt.


The existence of E[T(x)] is equivalent to the lim t( - tP') - O. Thus

ex == f tP,dt.

The complete-expectation-of-life at various ages is often used to compare levels of public health among different populations.


A similar integration by parts yields equivalent expressions for E[T(X)2]:


J I" tPy JL(x + t) at

o .

= 2 1: t tP, dt.

This result is useful in the calculation of Var [T(x)] by

Var[T(x)] - E[T(x)2] E[T(x)j2



Other characteristics of the distribution of T(x) can be determined. The median future lifetime of (x), to be denoted by 171(X), can be found by solving


Section 35 Other Life Table Charac

1 Pr[T(x) > I11(X)] = 2.


s[x + 111(X)] 1

-=- __ __;_:; =

sx 2


for II1(X). In particular, 111(0) is given by solving S[I11(O)] 1/2. We can also find

the mode of the distribution of T(x) by locating the value of t that yields a maximum

value of) x + t .

The expected value of K(x) is denoted bye, and is called the curtate-expectationof-life. By definition and use of summation by parts as described in Appendix 5, we have

r, = E[K] = L k kP, q., i:

k 0


Again, the existence of E[K(x)] is equivalent to the lim k( - kV,) = O. Thus, with a

change of the summation variable, k_x


e, = L w.. k~1


parts, we have

Following the outline used for tIle cOlitiuuous lIIudel and using summatJon by

2 ," 0

E[K(x) ] L. k- API ql I A


= L P l!.(-kP')



= F( - kP,)jO + L (l!.F)(k lIP.).



, The eXIstence i"!'£ E[K(x)2] is equivalent to lim F( kP') - O. With a change of the

summation variable, k-x

E[K(:t)2) Z (2k + 1) .-», 2.: (2k 1) kPx (3.5.9)


Var(K) = E[K2) - E[Kj2

= L (2k - 1) d\ - e;.



Chapter 3 Survival Distributions and Life Tables


To complete thr- dj~cussion of some of the entries in Table 3.3.1, we must define additional functions The symbol L, denotes the total expected number of years lived between llh('~' x and x + 1 by survivors of the initial group of I() lives. We have

L, = f t 1'+1 J.L(x + t) dt + 1'+1


where the integral counts the years lived of those who die between ages x and x + I, and tlte tell" l; I I COUIIts the years lIved between ages x and x + 1 by those

who survive to ilh(' .r + 1. Integration by parts yields -

L, = - E t dl'TI + I, ,1



The function t; i:-. a+so ((sed in definmg the central-death-rate over the interval from x to x + I, dunotcd by In, where

r 1"1 dt

_ l, 1'+1 L,


An application of this function is found in Chapter 10.

than one:

The definitions for Ill, and L, can be extended to age intervals of length other

"L, - )0 t I'TI J.L(x + f) dt + Izi.'+11


= [HI dt,

o .



//" v

f l'+1 J.L(x + t) dt

_ l, - 1'+11

f lX+1 dt ilL,


For the random survivorship group, "Lx is the total expected number of years lived

between ages x and .v + 11 by the survivors of the initial group of 10 lives and ,,111, is the average death rate experienced by tltis group over the mtcrval (x, x + n).

The symbol T, denotes the total number of years lived beyond age x by the survivorship group with III initial members. We have


Section 3.5 Other I ite TAble Characteristics

T, = J: t 1'41 IJ-(X + t) dt


1'-1 dt.


The final expression can be interpreted as the integral of the total time lived between ages x + t and x I t I dt by the lr+t lives who survive to that age interval.

We also recognize T, as the limit of "L, as 11 goes to infinity.

The average number of years of future lifetime of the I, survivors of the group at age x is given by

1,,",-1 dt

I, I,

as determined in (3.5.1) and (3.5.2).

We can express the average number of years lived between x and x + 11 by the I, survivors at age x as

X 11



= 0 IPx dt.


This function is the n-year temporary complete life expectancy of (x) and is denoted by ex:/il. (See Exercise 3.16.)

section, is the average number of years lived between ages x and x + 1 by those

A final function, rclated to the interpretation of the life table developed in this

of the survivorship group who die between those ages. This function is denoted by a(x) and is defined by

, .

, I

Chapter 3 Survival Distributions and Lite Tables


I , I

fl t 1'+1 f.l(X + t) dt


11(.1) = 1 •

f t.: f1(X + t) dt il


For the probabilistic view of the life table, we would have

t IV, f.l(X + t) dt

If we assume that

11+1 f.l(x + t) dt = d, dt

o ::::;: t ::::;: 1,

that is, if deaths are uniformly distributed in the year of age, we have

11(.1) =

where Figure 3.3.2 shows that the assumption may be inappropriate:

This is the 11s11al approximation for aU:), except for young and old years of age


Show that

L, = a(x) I, + [1 - a(x)] 1"1


L ::= I, + 1'+1.


o ution:

From (3.5.11), (3.5.12). and (3.5.18) we have

L - I

a(x) = x 1+1

I, - I'~l


L, = a(x) I, + [1 - 11(.1)] 1,~ l'

The formula

I, + 1+


can be JustIfIed by using the trapezoidal rule for approximate integration on (3.5.12). T

Key life table terminology, defined in Sections 3.3-3.5, is summarized as part of Table 3.9.1 in Section 3.9.


Section 3.5 Other Life Table Characteristics


Recursion Formulas

Example 3.5.1 illustrates !I~.!:!?_e of a numericalanalysistechnique to evaluate a life table characteristic. Ih~ trapezoidal .rule for approximate integration is .used. The calculation of complete and curtate expectations-of-life can be used to illustrate aI lother computational tool called recursion formulas. The application of recursion

formulas in this book typically involves one of two forms:

Backward Recursion Formula

u(x) = c(x) + d(x) u(x + 1)



Forward Recursion Formula

c(x) 1

u(x + 1) = - - + - u(x).

d(x) d(x)


Tire vatiable x is usually a JlolHLegative integer.

To evaluate a function u(x), for a domain of non-negative integer values ot x, we need to have available values of c(x) and d(x) and a starting value of u(x). This

procedure is used in subsequent chapters and is illustrated in Table 3.5.1 where backward recursion formulas are developed to compute e, and e,.

Backward Recursion Formulas for ex and ex

" 1. Basic eguation
3. Factor p, and
change variable in
the operation
4. Recursion
5. Startmg valuel I e, - J .P, ds


c, = r. + 2: ,p~ ,-2


e == x

,p, ds 1

e, = 11, + v, 2: ,PH I ,~I

= p, + /1, e,+1

ex = L .P. ds + v. r tV, I 1 dt = J: ,).1, ds + v, en1

u(x) = e; rex) = p,

U(x) = e" c(x) =

d(x) = p,

aThe integral crx) = OJ sp,ds can be evaluated using the trapezoidal rule as c(x) - (1 + p,)/2.

bFrom Section 3.3.1 we have Six) - 0, x ~ w, and srx) > 0, x < w. In this development we will assume that .~ is an integer.

Chapter 3 Survival Distributions and Life Tables


3.6 Assumptions for Fractional Ages

In this chapter we have discussed the continuous random variable remaining lifetime, T, and the discrete random variable curtate-future-lifetime, K. The life table developed in Section 3.3 specifies the probability distribution of K completely. To specify the distribution of T, we must postulate an analytic form or adopt a life

table and all assumption about the distribution between mtegers.

We wIll examine three assumptions that are widely used in actuarial science.

These will be stated in terms of the s.f. and in a form to show the nature of inter-

polation Over the interval (x, x + 1) implied by each assumption. In each statement, x is an integer and 0 ::::; t ::::; 1. The assumptions are the following:

• Linear interpolation: s(x + t) = (1 - t) s(.t) + t sex + 1). This is known as the uniform distribution or, perhaps more properly, a uniform distribution of deaths assumption within each year of age. Under this assumption tP\ is a linear function .

• Exponential interpolation, or linear interpolatiolt Oll log 5(.( I / ). log 5(.\ + l)

(1 - t) log sex) + t log S(X + 1). This is consistent with the assumption of a

constantjorce of mortalIty wIthm each year ot age. Under this assumption ill, is ex onential.

• Harmonic interpolation: 1/ sex + t) = (1 - t) / sex) + t / sex + 1). This is what is knovlH as the hyperbolic (historically Btllducci*) assumptioIt, fOl Ulldel it t!',


is a hyperbolic curve.

With these basic definitions, formulas can be derived for other standard probability functions in terms of life table probabilities. These results are presented in Table 3.6.1. Note that we just as well could have elected to propose equivalent definitions in terms of the p.d.f., the d.f., or the force of mortality.

The derivations of the entries in Table 3.6.1 are exercises in substituting the stated

assumption about sex + t) into the appropriate formulas of Sections 3.2' and 3.3. We will illustrate the process for the uniform distribution of deaths, an assumption

that is used extensively throughout this text.

To derive the first entry in the uniform distribution column, we start with sex) - sex + t)

tq, = sex)

o ~ t ~ I,

then substitute for sex + t),

sex) -[(1 - t) sex) + t sex + 1)] t [sex) - sex + 1)]

sex) sex)

For the second entry, we use (3.2.13) and

f-L(X + t) =

s'(x+t). sex + t) ,

"This assumption is named after G. Balducci, an Italian actuary, who pointed out its role in the traditional actuarial method of constructing life tables.


Section 3.6 Assumptions for Fractional Ages

Probability Theory Functions for Fractional Ages



Uniform Distribution

Constant Force



1 p,

1 - (1 - 1)11,

fJ.(x + f)


~Iog p,


(1 - f) II, 1 - III,


1 ~ fll,

(1 - I)q,

1 - p~


1 - (1 - Y - t)lI,

,p,fJ.(x + I)


v; log p,

, ,

[1 - (1 - t)II,j2

Note that, in this table, x is an integer, 0 < t < 1, 0 s: Y s: 1, Y + t s: 1, For rows one, three, four, and five, the relationships also hold for t = 0 and t = 1.

-then, substituting for sex + t), we have

__ ~[s~(x~")_-_s(~x_+_1~)] __ (x + t) =

1.1 [(1 - t) sex) + t sex + 1)] ,

Dividing both numerator and denominator of the right-hand side by sex) yields


The third entry is the special case of the fourth entry with y = 1 - t.

For the fourth entry we start with

_ sex + t) - sex + t + y)

yqHt - sex + t) r

then substitute for sex + t) and sex + t + y) to obtain

~/qx-+--t ==


t) s(x) + t s(x I 1)] [(1 t y) sex) I (t + y) sex + 1)]

(1 - t) sex) + t sex + 1)

l{[S(x) - sex + 1)] / sex)

{s(x) t[s(x) s(x + 1)]11 s(x)



1 - tq,'

The fifth entry is the complement of the first, and the final entry for the uniform distribution column is the product of the second and fifth entries.

Chapter 3 Survival Distributions and Life Tables


If, as before, x IS an integer, insight can be obtained by defining a random variable 5 = 5(x) by


(3.6.1 )

\ . . \~

where T is time-until-death, K is the curtate-future-lifetime, and 5 is the random variable representing the fractional part of a year lived in the year of death. Since K is a non-negative integer random variable and 5 is a continuous-type random

variable with all of its probability mass on the interval (0, 1), we can examine their joint distribution by writing

Pr [(K - k) n (5 <: s)] Pr (k < T <: k + 5)

= kJ.\ ,q,+k'

Now, using the expression for ,qHk under the uniform distribution assumption as shown in Table 3.6.1, we have

Pr [( K = k) n (5 ::::; 5)] = kP\ 5 q \ +]:

I , -

Pr(K- k) Pr(S < 5).


Therefore, the joint probability involving K and 5 can be factored into separate

probabilities of K and 5. It follows that, under the uniform distribution of deaths assumption, the random Variables K and S are independent. Since Pr (5 <: 5) S


Under the constant force of mortality assumption, are the random variables K and 5 independent?


Usmg entries from Table 3.6.1 for the constant force assumption, we obtain

Pr [(K k) I I (5 <: 5 )] AP, sq,. k

= kPx [1 - (p,+kn To discuss this result, we distinguish two cases:

• If PHk is not independent of k, we cannot factor the joint probability of K and 5 into separate probabilities. We conclude that K and 5 are not independent.

• In the special case where P,+k = P, a constant,

Pr [(K k) n (5 <: 5)] p:(1


(1 - pJp;(1 - p~)

(1 p,)

For this special case we conclude that K and 5 are independent under the constant force assumption. ..


Section 3.6 Assumptions for Fractional Ages


Under th~ assumption of uniform distribution of deaths, show that

o 1

a. ex = e, + "2

1 b. Var(T) = Var(K) + 12'



a. e,

= E[K] + E[S] 1

= e +-

x 2'

b. Var(T) = Var(K + 5).

From the independence of K and 5, under the uniform distribution assumption,

Var(T) = Var(K) + Vart S).

Further, since 5 is uniformly distributed over (0, 1),

1 Var(T) = Var(K) + -.

3.7 Some Analytical Laws of Mortality

\ There are three principal justifications for postulating an analytic form for mortality or survival functions. The first is philosophical. Many phenomena studied in physics can be explained efficiently by simple formulas Therefore, using biological

arguments, some authors have suggested that human survival is governed by an equally simple law. The secolLd justificatiolL is practical. It is easier to conUlLUlLicate

a function with a few parameters than it is to communicate a life table with perhaps

100 parameters or mortality probabilities. In addition, some of the analytic forms have elegant properties that are convenient in evaluating probability statements that involve more than one life. The third justification for a simple analytic survival function is the ease of estimating a few parameters of the function from mortality data.

The support for simple analytic survival functions has declined in recent years.

Many feel that the belief in universal laws of mortality is naive. With the increasing speed and storage capacity of compllter~, the advantages of some analytic forms

Nevertheless, some interesting research has recently reiterated the biological arguments for analytic laws of mortality.

In Table 3.7.1, several families of simple analytic mortality and survival functions, corresponding to various postulated laws, are displayed. The names of the originators of the laws and the dates of publication are included for identification purposes.

Chapter 3 Survival Distributions and Life Tables


Mortality and Survival Functions under Various Laws





De Moivre (1729)


x 1 --


exp] = nttc' - 1)]



B > 0, e > 1, x 2: °

A + Be"

exp] - Ax - /II(eX - 1)]

B > 0, A 2: .. B, c > 1, x 2: °


k > 0, /I > 0, X 2: °


• The special symbols are defined as


111 = --

log c'


II = ---

(II + 1)

• Gompertz's law is a special case of Makeham's law with A - o.

• If c 1 in Gompertz' sand Makeharrr' slaws, the exponentIal (constant force)

distribution results.

• In connection wIth Makeham's law, the constant A has been interpreted as capturing the accident hazard, and the term Be: as capturing the hazard of


The entries in the sex) column of Table 3.7.1 were obtained by substituting into (3.2.16). For example, for Makeham's law, we have

sex) = exp[-JCi(A + BC')ds]

exp [ Ax

log C

= exp] -Ax - 111 (e' - 1)]

were m B / log c.

Two objectives governed the development of a mortality table for computational purposes in the examples and exercises of this book. One objective was to have mortality rates in the middle of the range of variation for groups, such variation caused by factors such as residence, gender, insured status, annuity status, marital status, and occupation The second objective was to ha v e a Makeham law at most

ages to illustrate how calculations for multiple lives can be performed.

ages 13 and grea ter,

The Illustrative Life Table in Appendix 2A is based on the Makeham law for

1,000 j.L(x) = 0.7 + 0.05 (l0004)'.

(3.7.1 )

Section 3.7 Some Analytical Laws of Mortality

The calculations of the basic fUnctiol1S ,/" t; dilL! ii, from (3.7.1) were all done

directly from (3.7.1) instead of calculating I, and d; from the truncated values of t],. It was found that the latter choice would make little difference in the applications. It should be kept in mind that the Illustrative Life Table, as its name implies, is for illustrative purposes only.

3.8 Select and Ultimate Tables

In Section 3.2 we discussed how 1)1, [the probability that (x) will survive to age :t + /] l1IigItt be illtelpreted ill t" 0 ways. TIte first illterpletatioll was that tIte

probability can be evaluated by a survival function appropriate for newborns, under the single hypothesis that the newborn has survived to age .Y. This interpretation has been the basis of the notation and development of the formulas. The second interpretation was that additional knowledge available about the life at age x might make the original survival function inappropriate for evaluating probability statements about the future lifetime of (x). For example, the life might have been un-


derwritten and accepted for life insurance at age x. This information would lead us to believe that (x)'s future-lifetime distribution is different from what we might

otherwise assume for lives age x. As a second example, the life might have become disabled at age r: This illformatioll would lead us to belie\e that tIte future-iifetime

distribution for (x) is different from that of those not disabled at age x. In these

two illustrations, a special force of mortality that incorporates the particular information available at age x would be preferred. Without this particular information for (x), the form of mortality at duration t would be a function of only the attained a~e x + t, denoted in the previous sections by f.l(x + t). Given the additional information at x, the force of mortality at x + t is a function of this information at x and duration t. Its notation will be f.lAt), showing separately the age, x, at which the additional information was available, and the duration, t, The additional infor-

. mation is usually not explicit in the notation but is conveyed by the context. In other words, tIle complete model for such lives is a set of survival functions in

eluding one for each age at which information is available on issue of insurance,

disability, and so on. Il1ii~~_t ofsurvival functions can be thought of as a function

. __ of two variables. One variable is the age at selection (e.g., at policy issue or the onset of disability), [x), and the second variable is the duration since policy issue or duration since selection, t. Then each of the usual life table functions associated with this bivariate survival function is a two-dimensional array on [x) and t. Note the bracket notation to indicate which variable identifies the age of selection.

The schematic diagram in Figure 3.8.1 illustrates these ideas. For instance, suppose some special information is available about a group of lives age 30. Perhaps

they have been accepted for life insurance or perhaps they have become disabled.

A speCIal lIte table can be bUllt for these lIves. The condItIonal probabIlIty of death in each year of duration would be denoted by Q]30]+i' i = 0, 1, 2, ... , and would be entered on the first row of Figure 3.8.1. The subscript reflects the bivariate nature of this function with the bracketed thirty, [30], denoting that the survival function in the first row is conditional on special information available at age 30. The second row of Figure 3.8.1 would contain the probabilities of death for lives on which the

Chapter 3 Survival Distributions and Life Tables


________________________ ... -_.--- .... _.= .... =- =.~.-==.-.=. =. -=_··_--c·,,'_-_ -------. -~--.-------- ~_~